4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/ofd/ofd_grant.c
38 * Author: Johann Lombardi <johann@whamcloud..com>
41 #define DEBUG_SUBSYSTEM S_FILTER
43 #include "ofd_internal.h"
45 /* At least enough to send a couple of 1MB RPCs, even if not max sized */
46 #define OFD_GRANT_CHUNK (2ULL * DT_MAX_BRW_SIZE)
48 /* Clients typically hold 2x their max_rpcs_in_flight of grant space */
49 #define OFD_GRANT_SHRINK_LIMIT(exp) (2ULL * 8 * exp_max_brw_size(exp))
51 static inline obd_size ofd_grant_from_cli(struct obd_export *exp,
52 struct ofd_device *ofd, obd_size val)
54 if (ofd_grant_compat(exp, ofd))
55 /* clients not supporting OBD_CONNECT_GRANT_PARAM actually
56 * consume 4KB of grant per block, we should thus inflate
57 * the grant counters to reflect what was actually consumed */
58 return val << (ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT);
62 static inline obd_size ofd_grant_to_cli(struct obd_export *exp,
63 struct ofd_device *ofd, obd_size val)
65 if (ofd_grant_compat(exp, ofd))
66 return val >> (ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT);
70 static inline obd_size ofd_grant_chunk(struct obd_export *exp,
71 struct ofd_device *ofd)
73 if (ofd_obd(ofd)->obd_self_export == exp)
74 /* Grant enough space to handle a big precreate request */
75 return OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
77 if (ofd_grant_compat(exp, ofd))
78 /* Try to grant enough space to send a full-size RPC */
79 return exp_max_brw_size(exp) <<
80 (ofd->ofd_blockbits - COMPAT_BSIZE_SHIFT);
82 /* Try to return enough to send two full RPCs, if needed */
83 return exp_max_brw_size(exp) * 2;
87 * Perform extra sanity checks for grant accounting. This is done at connect,
88 * disconnect, and statfs RPC time, so it shouldn't be too bad. We can
89 * always get rid of it or turn it off when we know accounting is good.
91 * \param obd - is the device to check
92 * \param func - is the function to call if an inconsistency is found
94 void ofd_grant_sanity_check(struct obd_device *obd, const char *func)
96 struct filter_export_data *fed;
97 struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
98 struct obd_export *exp;
100 obd_size tot_dirty = 0;
101 obd_size tot_pending = 0;
102 obd_size tot_granted = 0;
103 obd_size fo_tot_dirty, fo_tot_pending;
104 obd_size fo_tot_granted;
106 if (cfs_list_empty(&obd->obd_exports))
109 /* We don't want to do this for large machines that do lots of
110 * mounts or unmounts. It burns... */
111 if (obd->obd_num_exports > 100)
114 maxsize = ofd->ofd_osfs.os_blocks << ofd->ofd_blockbits;
116 spin_lock(&obd->obd_dev_lock);
117 spin_lock(&ofd->ofd_grant_lock);
118 cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
121 fed = &exp->exp_filter_data;
123 if (obd->obd_self_export == exp)
124 CDEBUG(D_CACHE, "%s: processing self export: %ld %ld "
125 "%ld\n", obd->obd_name, fed->fed_grant,
126 fed->fed_pending, fed->fed_dirty);
128 if (fed->fed_grant < 0 || fed->fed_pending < 0 ||
131 if (fed->fed_grant + fed->fed_pending > maxsize) {
132 CERROR("%s: cli %s/%p fed_grant(%ld) + fed_pending(%ld)"
133 " > maxsize("LPU64")\n", obd->obd_name,
134 exp->exp_client_uuid.uuid, exp, fed->fed_grant,
135 fed->fed_pending, maxsize);
136 spin_unlock(&obd->obd_dev_lock);
137 spin_unlock(&ofd->ofd_grant_lock);
140 if (fed->fed_dirty > maxsize) {
141 CERROR("%s: cli %s/%p fed_dirty(%ld) > maxsize("LPU64
142 ")\n", obd->obd_name, exp->exp_client_uuid.uuid,
143 exp, fed->fed_dirty, maxsize);
144 spin_unlock(&obd->obd_dev_lock);
145 spin_unlock(&ofd->ofd_grant_lock);
148 CDEBUG_LIMIT(error ? D_ERROR : D_CACHE, "%s: cli %s/%p dirty "
149 "%ld pend %ld grant %ld\n", obd->obd_name,
150 exp->exp_client_uuid.uuid, exp, fed->fed_dirty,
151 fed->fed_pending, fed->fed_grant);
152 tot_granted += fed->fed_grant + fed->fed_pending;
153 tot_pending += fed->fed_pending;
154 tot_dirty += fed->fed_dirty;
156 spin_unlock(&obd->obd_dev_lock);
157 fo_tot_granted = ofd->ofd_tot_granted;
158 fo_tot_pending = ofd->ofd_tot_pending;
159 fo_tot_dirty = ofd->ofd_tot_dirty;
161 if (tot_granted != fo_tot_granted)
162 CERROR("%s: tot_granted "LPU64" != fo_tot_granted "LPU64"\n",
163 func, tot_granted, fo_tot_granted);
164 if (tot_pending != fo_tot_pending)
165 CERROR("%s: tot_pending "LPU64" != fo_tot_pending "LPU64"\n",
166 func, tot_pending, fo_tot_pending);
167 if (tot_dirty != fo_tot_dirty)
168 CERROR("%s: tot_dirty "LPU64" != fo_tot_dirty "LPU64"\n",
169 func, tot_dirty, fo_tot_dirty);
170 if (tot_pending > tot_granted)
171 CERROR("%s: tot_pending "LPU64" > tot_granted "LPU64"\n",
172 func, tot_pending, tot_granted);
173 if (tot_granted > maxsize)
174 CERROR("%s: tot_granted "LPU64" > maxsize "LPU64"\n",
175 func, tot_granted, maxsize);
176 if (tot_dirty > maxsize)
177 CERROR("%s: tot_dirty "LPU64" > maxsize "LPU64"\n",
178 func, tot_dirty, maxsize);
179 spin_unlock(&ofd->ofd_grant_lock);
183 * Get fresh statfs information from the OSD layer if the cache is older than 1s
184 * or if force is set. The OSD layer is in charge of estimating data & metadata
187 * \param env - is the lu environment passed by the caller
188 * \param exp - export used to print client info in debug messages
189 * \param force - is used to force a refresh of statfs information
190 * \param from_cache - returns whether the statfs information are
193 static void ofd_grant_statfs(const struct lu_env *env, struct obd_export *exp,
194 int force, int *from_cache)
196 struct obd_device *obd = exp->exp_obd;
197 struct ofd_device *ofd = ofd_exp(exp);
198 struct obd_statfs *osfs = &ofd_info(env)->fti_u.osfs;
203 max_age = 0; /* get fresh statfs data */
205 max_age = cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS);
207 rc = ofd_statfs_internal(env, ofd, osfs, max_age, from_cache);
214 CDEBUG(D_CACHE, "%s: cli %s/%p free: "LPU64" avail: "LPU64"\n",
215 obd->obd_name, exp->exp_client_uuid.uuid, exp,
216 osfs->os_bfree << ofd->ofd_blockbits,
217 osfs->os_bavail << ofd->ofd_blockbits);
221 * Figure out how much space is available on the backend filesystem.
222 * This is done by accessing cached statfs data previously populated by
223 * ofd_grant_statfs(), from which we withdraw the space already granted to
224 * clients and the reserved space.
226 * \param exp - export which received the write request
228 static obd_size ofd_grant_space_left(struct obd_export *exp)
230 struct obd_device *obd = exp->exp_obd;
231 struct ofd_device *ofd = ofd_exp(exp);
232 obd_size tot_granted;
233 obd_size left, avail;
237 LASSERT(spin_is_locked(&ofd->ofd_grant_lock));
239 spin_lock(&ofd->ofd_osfs_lock);
240 /* get available space from cached statfs data */
241 left = ofd->ofd_osfs.os_bavail << ofd->ofd_blockbits;
242 unstable = ofd->ofd_osfs_unstable; /* those might be accounted twice */
243 spin_unlock(&ofd->ofd_osfs_lock);
245 tot_granted = ofd->ofd_tot_granted;
247 if (left < tot_granted) {
248 int mask = (left + unstable <
249 tot_granted - ofd->ofd_tot_pending) ?
252 CDEBUG_LIMIT(mask, "%s: cli %s/%p left "LPU64" < tot_grant "
253 LPU64" unstable "LPU64" pending "LPU64" "
255 obd->obd_name, exp->exp_client_uuid.uuid, exp,
256 left, tot_granted, unstable,
257 ofd->ofd_tot_pending, ofd->ofd_tot_dirty);
262 /* Withdraw space already granted to clients */
265 /* If the left space is below the grant threshold x available space,
266 * stop granting space to clients.
267 * The purpose of this threshold is to keep some error margin on the
268 * overhead estimate made by the OSD layer. If we grant all the free
269 * space, we have no way (grant space cannot be revoked yet) to
270 * adjust if the write overhead has been underestimated. */
271 left -= min_t(obd_size, left, ofd_grant_reserved(ofd, avail));
273 /* Align left on block size */
274 left &= ~((1ULL << ofd->ofd_blockbits) - 1);
276 CDEBUG(D_CACHE, "%s: cli %s/%p avail "LPU64" left "LPU64" unstable "
277 LPU64" tot_grant "LPU64" pending "LPU64"\n", obd->obd_name,
278 exp->exp_client_uuid.uuid, exp, avail, left, unstable,
279 tot_granted, ofd->ofd_tot_pending);
285 * Grab the dirty and seen grant announcements from the incoming obdo.
286 * We will later calculate the client's new grant and return it.
287 * Caller must hold ofd_grant_lock spinlock.
289 * \param env - is the lu environment supplying osfs storage
290 * \param exp - is the export for which we received the request
291 * \paral oa - is the incoming obdo sent by the client
293 static void ofd_grant_incoming(const struct lu_env *env, struct obd_export *exp,
296 struct filter_export_data *fed;
297 struct ofd_device *ofd = ofd_exp(exp);
298 struct obd_device *obd = exp->exp_obd;
299 long dirty, dropped, grant_chunk;
302 LASSERT(spin_is_locked(&ofd->ofd_grant_lock));
304 if ((oa->o_valid & (OBD_MD_FLBLOCKS|OBD_MD_FLGRANT)) !=
305 (OBD_MD_FLBLOCKS|OBD_MD_FLGRANT)) {
306 oa->o_valid &= ~OBD_MD_FLGRANT;
310 fed = &exp->exp_filter_data;
312 /* Add some margin, since there is a small race if other RPCs arrive
313 * out-or-order and have already consumed some grant. We want to
314 * leave this here in case there is a large error in accounting. */
316 "%s: cli %s/%p reports grant "LPU64" dropped %u, local %lu\n",
317 obd->obd_name, exp->exp_client_uuid.uuid, exp, oa->o_grant,
318 oa->o_dropped, fed->fed_grant);
320 if ((long long)oa->o_dirty < 0)
323 dirty = ofd_grant_from_cli(exp, ofd, oa->o_dirty);
324 dropped = ofd_grant_from_cli(exp, ofd, (obd_size)oa->o_dropped);
325 grant_chunk = ofd_grant_chunk(exp, ofd);
327 /* Update our accounting now so that statfs takes it into account.
328 * Note that fed_dirty is only approximate and can become incorrect
329 * if RPCs arrive out-of-order. No important calculations depend
330 * on fed_dirty however, but we must check sanity to not assert. */
331 if (dirty > fed->fed_grant + 4 * grant_chunk)
332 dirty = fed->fed_grant + 4 * grant_chunk;
333 ofd->ofd_tot_dirty += dirty - fed->fed_dirty;
334 if (fed->fed_grant < dropped) {
336 "%s: cli %s/%p reports %lu dropped > grant %lu\n",
337 obd->obd_name, exp->exp_client_uuid.uuid, exp, dropped,
341 if (ofd->ofd_tot_granted < dropped) {
342 CERROR("%s: cli %s/%p reports %lu dropped > tot_grant "LPU64
343 "\n", obd->obd_name, exp->exp_client_uuid.uuid, exp,
344 dropped, ofd->ofd_tot_granted);
347 ofd->ofd_tot_granted -= dropped;
348 fed->fed_grant -= dropped;
349 fed->fed_dirty = dirty;
351 if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0) {
352 CERROR("%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
353 obd->obd_name, exp->exp_client_uuid.uuid, exp,
354 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
355 spin_unlock(&ofd->ofd_grant_lock);
362 * Called when the client is able to release some grants. Proceed with the
363 * shrink request when there is less ungranted space remaining
364 * than the amount all of the connected clients would consume if they
365 * used their full grant.
367 * \param exp - is the export for which we received the request
368 * \paral oa - is the incoming obdo sent by the client
369 * \param left_space - is the remaining free space with space already granted
372 static void ofd_grant_shrink(struct obd_export *exp,
373 struct obdo *oa, obd_size left_space)
375 struct filter_export_data *fed;
376 struct ofd_device *ofd = ofd_exp(exp);
377 struct obd_device *obd = exp->exp_obd;
380 LASSERT(spin_is_locked(&ofd->ofd_grant_lock));
382 if (left_space >= ofd->ofd_tot_granted_clients *
383 OFD_GRANT_SHRINK_LIMIT(exp))
386 grant_shrink = ofd_grant_from_cli(exp, ofd, oa->o_grant);
388 fed = &exp->exp_filter_data;
389 fed->fed_grant -= grant_shrink;
390 ofd->ofd_tot_granted -= grant_shrink;
392 CDEBUG(D_CACHE, "%s: cli %s/%p shrink %ld fed_grant %ld total "
393 LPU64"\n", obd->obd_name, exp->exp_client_uuid.uuid,
394 exp, grant_shrink, fed->fed_grant, ofd->ofd_tot_granted);
396 /* client has just released some grant, don't grant any space back */
401 * Calculate how much space is required to write a given network buffer
403 static inline int ofd_grant_rnb_size(struct obd_export *exp,
404 struct ofd_device *ofd,
405 struct niobuf_remote *rnb)
407 obd_size blocksize, bytes, end;
409 if (exp && ofd_grant_compat(exp, ofd))
410 blocksize = 1ULL << COMPAT_BSIZE_SHIFT;
412 blocksize = 1ULL << ofd->ofd_blockbits;
414 /* The network buffer might span several blocks, align it on block
416 bytes = rnb->rnb_offset & (blocksize - 1);
417 bytes += rnb->rnb_len;
418 end = bytes & (blocksize - 1);
420 bytes += blocksize - end;
422 /* Apply per-export pecularities if one is given */
423 bytes = ofd_grant_from_cli(exp, ofd, (obd_size)bytes);
429 * When clients have dirtied as much space as they've been granted they
430 * fall through to sync writes. These sync writes haven't been expressed
431 * in grants and need to error with ENOSPC when there isn't room in the
432 * filesystem for them after grants are taken into account. However,
433 * writeback of the dirty data that was already granted space can write
435 * Caller must hold ofd_grant_lock spinlock.
437 * \param env - is the lu environment passed by the caller
438 * \param exp - is the export identifying the client which sent the RPC
439 * \param oa - is the incoming obdo in which we should return the pack the
441 * \param rnb - is the list of network buffers
442 * \param niocont - is the number of network buffers in the list
443 * \param left - is the remaining free space with space already granted
446 static void ofd_grant_check(const struct lu_env *env, struct obd_export *exp,
447 struct obdo *oa, struct niobuf_remote *rnb,
448 int niocount, obd_size *left)
450 struct filter_export_data *fed = &exp->exp_filter_data;
451 struct obd_device *obd = exp->exp_obd;
452 struct ofd_device *ofd = ofd_exp(exp);
453 unsigned long ungranted = 0;
454 unsigned long granted = 0;
457 struct ofd_thread_info *info = ofd_info(env);
461 LASSERT(spin_is_locked(&ofd->ofd_grant_lock));
463 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
464 (oa->o_flags & OBD_FL_RECOV_RESEND)) {
466 CDEBUG(D_CACHE, "Recoverable resend arrived, skipping "
470 for (i = 0; i < niocount; i++) {
473 if (obd->obd_recovering) {
474 /* Replaying write. Grant info have been processed
475 * already so no need to do any enforcement here.
476 * It is worth noting that only bulk writes with all
477 * rnbs having OBD_BRW_FROM_GRANT can be replayed.
478 * If one page hasn't OBD_BRW_FROM_GRANT set, then
479 * the whole bulk is written synchronously */
480 if (rnb[i].rnb_flags & OBD_BRW_FROM_GRANT) {
481 rnb[i].rnb_flags |= OBD_BRW_GRANTED;
484 CERROR("%s: cli %s is replaying OST_WRITE "
485 "while one rnb hasn't OBD_BRW_FROM_GRANT"
486 " set (0x%x)\n", exp->exp_obd->obd_name,
487 exp->exp_client_uuid.uuid,
491 } else if ((oa->o_valid & OBD_MD_FLGRANT) &&
492 (rnb[i].rnb_flags & OBD_BRW_FROM_GRANT)) {
494 /* This is a recoverable resend so grant
495 * information have already been processed */
496 rnb[i].rnb_flags |= OBD_BRW_GRANTED;
500 /* inflate consumed space if needed */
501 bytes = ofd_grant_rnb_size(exp, ofd, &rnb[i]);
502 if (fed->fed_grant < granted + bytes) {
503 CDEBUG(D_CACHE, "%s: cli %s/%p claims %ld+%d "
504 "GRANT, real grant %lu idx %d\n",
505 exp->exp_obd->obd_name,
506 exp->exp_client_uuid.uuid, exp,
507 granted, bytes, fed->fed_grant, i);
510 rnb[i].rnb_flags |= OBD_BRW_GRANTED;
515 /* Consume grant space on the server.
516 * Unlike above, ofd_grant_rnb_size() is called with exp = NULL
517 * so that the required grant space isn't inflated. This is
518 * done on purpose since the server can deal with large block
519 * size, unlike some clients */
520 bytes = ofd_grant_rnb_size(NULL, ofd, &rnb[i]);
521 if (*left > ungranted + bytes) {
522 /* if enough space, pretend it was granted */
524 rnb[i].rnb_flags |= OBD_BRW_GRANTED;
528 /* We can't check for already-mapped blocks here (make sense
529 * when backend filesystem does not use COW) as it requires
530 * dropping the grant lock.
531 * Instead, we clear ~OBD_BRW_GRANTED and in that case we need
532 * to go through and verify if all of the blocks not marked
533 * BRW_GRANTED are already mapped and we can ignore this error.
535 rnb[i].rnb_flags &= ~OBD_BRW_GRANTED;
536 CDEBUG(D_CACHE,"%s: cli %s/%p idx %d no space for %d\n",
537 exp->exp_obd->obd_name,
538 exp->exp_client_uuid.uuid, exp, i, bytes);
541 /* record space used for the I/O, will be used in ofd_grant_commmit() */
542 /* Now substract what the clients has used already. We don't subtract
543 * this from the tot_granted yet, so that other client's can't grab
544 * that space before we have actually allocated our blocks. That
545 * happens in ofd_grant_commit() after the writes are done. */
546 info->fti_used = granted + ungranted;
548 fed->fed_grant -= granted;
549 fed->fed_pending += info->fti_used;
550 ofd->ofd_tot_granted += ungranted;
551 ofd->ofd_tot_pending += info->fti_used;
554 "%s: cli %s/%p granted: %lu ungranted: %lu grant: %lu dirty: %lu"
555 "\n", obd->obd_name, exp->exp_client_uuid.uuid, exp,
556 granted, ungranted, fed->fed_grant, fed->fed_dirty);
558 if (obd->obd_recovering)
559 /* don't update dirty accounting during recovery */
562 if (fed->fed_dirty < granted) {
563 CWARN("%s: cli %s/%p claims granted %lu > fed_dirty %lu\n",
564 obd->obd_name, exp->exp_client_uuid.uuid, exp,
565 granted, fed->fed_dirty);
566 granted = fed->fed_dirty;
568 ofd->ofd_tot_dirty -= granted;
569 fed->fed_dirty -= granted;
571 if (fed->fed_dirty < 0 || fed->fed_grant < 0 || fed->fed_pending < 0) {
572 CERROR("%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
573 obd->obd_name, exp->exp_client_uuid.uuid, exp,
574 fed->fed_dirty, fed->fed_pending, fed->fed_grant);
575 spin_unlock(&ofd->ofd_grant_lock);
582 * Calculate how much grant space to return to client, based on how much space
583 * is currently free and how much of that is already granted.
584 * Caller must hold ofd_grant_lock spinlock.
586 * \param exp - is the export of the client which sent the request
587 * \param curgrant - is the current grant claimed by the client
588 * \param want - is how much grant space the client would like to have
589 * \param left - is the remaining free space with granted space taken out
590 * \param conservative - is how server grants, if true, a certain amount, else
591 * server will grant as client requested.
593 static long ofd_grant(struct obd_export *exp, obd_size curgrant,
594 obd_size want, obd_size left, bool conservative)
596 struct obd_device *obd = exp->exp_obd;
597 struct ofd_device *ofd = ofd_exp(exp);
598 struct filter_export_data *fed = &exp->exp_filter_data;
604 if (ofd_grant_prohibit(exp, ofd) || left == 0 || exp->exp_failed)
607 if (want > 0x7fffffff) {
608 CERROR("%s: client %s/%p requesting > 2GB grant "LPU64"\n",
609 obd->obd_name, exp->exp_client_uuid.uuid, exp, want);
613 /* client not supporting OBD_CONNECT_GRANT_PARAM works with a 4KB block
614 * size while the reality is different */
615 curgrant = ofd_grant_from_cli(exp, ofd, curgrant);
616 want = ofd_grant_from_cli(exp, ofd, want);
617 grant_chunk = ofd_grant_chunk(exp, ofd);
619 /* Grant some fraction of the client's requested grant space so that
620 * they are not always waiting for write credits (not all of it to
621 * avoid overgranting in face of multiple RPCs in flight). This
622 * essentially will be able to control the OSC_MAX_RIF for a client.
624 * If we do have a large disparity between what the client thinks it
625 * has and what we think it has, don't grant very much and let the
626 * client consume its grant first. Either it just has lots of RPCs
627 * in flight, or it was evicted and its grants will soon be used up. */
628 if (curgrant >= want || curgrant >= fed->fed_grant + grant_chunk)
631 if (obd->obd_recovering)
632 conservative = false;
635 /* don't grant more than 1/8th of the remaining free space in
638 grant = min(want, left);
639 /* round grant upt to the next block size */
640 grant = (grant + (1 << ofd->ofd_blockbits) - 1) &
641 ~((1ULL << ofd->ofd_blockbits) - 1);
646 /* Limit to ofd_grant_chunk() if not reconnect/recovery */
647 if ((grant > grant_chunk) && conservative)
650 ofd->ofd_tot_granted += grant;
651 fed->fed_grant += grant;
653 if (fed->fed_grant < 0) {
654 CERROR("%s: cli %s/%p grant %ld want "LPU64" current "LPU64"\n",
655 obd->obd_name, exp->exp_client_uuid.uuid, exp,
656 fed->fed_grant, want, curgrant);
657 spin_unlock(&ofd->ofd_grant_lock);
662 "%s: cli %s/%p wants: "LPU64" current grant "LPU64
663 " granting: "LPU64"\n", obd->obd_name, exp->exp_client_uuid.uuid,
664 exp, want, curgrant, grant);
666 "%s: cli %s/%p tot cached:"LPU64" granted:"LPU64
667 " num_exports: %d\n", obd->obd_name, exp->exp_client_uuid.uuid,
668 exp, ofd->ofd_tot_dirty, ofd->ofd_tot_granted,
669 obd->obd_num_exports);
671 RETURN(ofd_grant_to_cli(exp, ofd, grant));
675 * Client connection or reconnection.
677 * \param env - is the lu environment provided by the caller
678 * \param exp - is the client's export which is reconnecting
679 * \param want - is how much the client would like to get
680 * \param conservative - is how server grants to client, if true server will
681 * only grant certain amount, else server will grant client requested
684 long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
685 obd_size want, bool conservative)
687 struct ofd_device *ofd = ofd_exp(exp);
688 struct filter_export_data *fed = &exp->exp_filter_data;
692 int force = 0; /* can use cached data */
694 /* don't grant space to client with read-only access */
695 if ((exp_connect_flags(exp) & OBD_CONNECT_RDONLY) ||
696 ofd_grant_prohibit(exp, ofd))
700 ofd_grant_statfs(env, exp, force, &from_cache);
702 spin_lock(&ofd->ofd_grant_lock);
704 /* Grab free space from cached info and take out space already granted
705 * to clients as well as reserved space */
706 left = ofd_grant_space_left(exp);
708 /* get fresh statfs data if we are short in ungranted space */
709 if (from_cache && left < 32 * ofd_grant_chunk(exp, ofd)) {
710 spin_unlock(&ofd->ofd_grant_lock);
711 CDEBUG(D_CACHE, "fs has no space left and statfs too old\n");
716 ofd_grant(exp, ofd_grant_to_cli(exp, ofd, (obd_size)fed->fed_grant),
717 want, left, conservative);
719 /* return to client its current grant */
720 grant = ofd_grant_to_cli(exp, ofd, (obd_size)fed->fed_grant);
721 ofd->ofd_tot_granted_clients++;
723 spin_unlock(&ofd->ofd_grant_lock);
725 CDEBUG(D_CACHE, "%s: cli %s/%p ocd_grant: %ld want: "LPU64" left: "
726 LPU64"\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
727 exp, grant, want, left);
733 * Remove a client from the grant accounting totals. We also remove
734 * the export from the obd device under the osfs and dev locks to ensure
735 * that the ofd_grant_sanity_check() calculations are always valid.
736 * The client should do something similar when it invalidates its import.
738 * \param exp - is the client's export to remove from grant accounting
740 void ofd_grant_discard(struct obd_export *exp)
742 struct obd_device *obd = exp->exp_obd;
743 struct ofd_device *ofd = ofd_exp(exp);
744 struct filter_export_data *fed = &exp->exp_filter_data;
746 spin_lock(&ofd->ofd_grant_lock);
747 LASSERTF(ofd->ofd_tot_granted >= fed->fed_grant,
748 "%s: tot_granted "LPU64" cli %s/%p fed_grant %ld\n",
749 obd->obd_name, ofd->ofd_tot_granted,
750 exp->exp_client_uuid.uuid, exp, fed->fed_grant);
751 ofd->ofd_tot_granted -= fed->fed_grant;
753 LASSERTF(ofd->ofd_tot_pending >= fed->fed_pending,
754 "%s: tot_pending "LPU64" cli %s/%p fed_pending %ld\n",
755 obd->obd_name, ofd->ofd_tot_pending,
756 exp->exp_client_uuid.uuid, exp, fed->fed_pending);
757 /* ofd_tot_pending is handled in ofd_grant_commit as bulk
759 LASSERTF(ofd->ofd_tot_dirty >= fed->fed_dirty,
760 "%s: tot_dirty "LPU64" cli %s/%p fed_dirty %ld\n",
761 obd->obd_name, ofd->ofd_tot_dirty,
762 exp->exp_client_uuid.uuid, exp, fed->fed_dirty);
763 ofd->ofd_tot_dirty -= fed->fed_dirty;
765 spin_unlock(&ofd->ofd_grant_lock);
769 * Called at prepare time when handling read request. This function extracts
770 * incoming grant information from the obdo and processes the grant shrink
773 * \param env - is the lu environment provided by the caller
774 * \param exp - is the export of the client which sent the request
775 * \paral oa - is the incoming obdo sent by the client
777 void ofd_grant_prepare_read(const struct lu_env *env,
778 struct obd_export *exp, struct obdo *oa)
780 struct ofd_device *ofd = ofd_exp(exp);
787 if ((oa->o_valid & OBD_MD_FLGRANT) == 0)
788 /* The read request does not contain any grant
792 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
793 (oa->o_flags & OBD_FL_SHRINK_GRANT)) {
794 /* To process grant shrink request, we need to know how much
795 * available space remains on the backend filesystem.
796 * Shrink requests are not so common, we always get fresh
797 * statfs information. */
798 ofd_grant_statfs(env, exp, 1, NULL);
800 /* protect all grant counters */
801 spin_lock(&ofd->ofd_grant_lock);
803 /* Grab free space from cached statfs data and take out space
804 * already granted to clients as well as reserved space */
805 left = ofd_grant_space_left(exp);
807 /* all set now to proceed with shrinking */
810 /* no grant shrinking request packed in the obdo and
811 * since we don't grant space back on reads, no point
812 * in running statfs, so just skip it and process
813 * incoming grant data directly. */
814 spin_lock(&ofd->ofd_grant_lock);
818 /* extract incoming grant infomation provided by the client */
819 ofd_grant_incoming(env, exp, oa);
821 /* unlike writes, we don't return grants back on reads unless a grant
822 * shrink request was packed and we decided to turn it down. */
824 ofd_grant_shrink(exp, oa, left);
828 spin_unlock(&ofd->ofd_grant_lock);
832 * Called at write prepare time to handle incoming grant, check that we have
833 * enough space and grant some space back to the client if possible.
835 * \param env - is the lu environment provided by the caller
836 * \param exp - is the export of the client which sent the request
837 * \paral oa - is the incoming obdo sent by the client
838 * \param rnb - is the list of network buffers
839 * \param niocont - is the number of network buffers in the list
841 void ofd_grant_prepare_write(const struct lu_env *env,
842 struct obd_export *exp, struct obdo *oa,
843 struct niobuf_remote *rnb, int niocount)
845 struct obd_device *obd = exp->exp_obd;
846 struct ofd_device *ofd = ofd_exp(exp);
849 int force = 0; /* can use cached data intially */
855 /* get statfs information from OSD layer */
856 ofd_grant_statfs(env, exp, force, &from_cache);
858 spin_lock(&ofd->ofd_grant_lock); /* protect all grant counters */
860 /* Grab free space from cached statfs data and take out space already
861 * granted to clients as well as reserved space */
862 left = ofd_grant_space_left(exp);
864 /* Get fresh statfs data if we are short in ungranted space */
865 if (from_cache && left < 32 * ofd_grant_chunk(exp, ofd)) {
866 spin_unlock(&ofd->ofd_grant_lock);
867 CDEBUG(D_CACHE, "%s: fs has no space left and statfs too old\n",
873 /* When close to free space exhaustion, trigger a sync to force
874 * writeback cache to consume required space immediately and release as
875 * much space as possible. */
876 if (!obd->obd_recovering && force != 2 && left < OFD_GRANT_CHUNK) {
877 bool from_grant = true;
880 /* That said, it is worth running a sync only if some pages did
881 * not consume grant space on the client and could thus fail
882 * with ENOSPC later in ofd_grant_check() */
883 for (i = 0; i < niocount; i++)
884 if (!(rnb[i].rnb_flags & OBD_BRW_FROM_GRANT))
888 /* at least one network buffer requires acquiring grant
889 * space on the server */
890 spin_unlock(&ofd->ofd_grant_lock);
891 /* discard errors, at least we tried ... */
892 rc = dt_sync(env, ofd->ofd_osd);
898 /* extract incoming grant information provided by the client */
899 ofd_grant_incoming(env, exp, oa);
902 ofd_grant_check(env, exp, oa, rnb, niocount, &left);
904 if (!(oa->o_valid & OBD_MD_FLGRANT)) {
905 spin_unlock(&ofd->ofd_grant_lock);
909 /* if OBD_FL_SHRINK_GRANT is set, the client is willing to release some
911 if ((oa->o_valid & OBD_MD_FLFLAGS) &&
912 (oa->o_flags & OBD_FL_SHRINK_GRANT))
913 ofd_grant_shrink(exp, oa, left);
915 /* grant more space back to the client if possible */
916 oa->o_grant = ofd_grant(exp, oa->o_grant, oa->o_undirty, left,
918 spin_unlock(&ofd->ofd_grant_lock);
922 * Called during object precreation to consume grant space.
923 * More space is granted for precreation if possible.
925 * \param env - is the lu environment provided by the caller
926 * \param exp - is the export holding the grant space for precreation (= self
928 * \paral nr - is the number of objects the caller wants to create objects
930 int ofd_grant_create(const struct lu_env *env, struct obd_export *exp, int *nr)
932 struct ofd_thread_info *info = ofd_info(env);
933 struct ofd_device *ofd = ofd_exp(exp);
934 struct filter_export_data *fed = &exp->exp_filter_data;
936 unsigned long wanted;
941 if (exp->exp_obd->obd_recovering ||
942 ofd->ofd_dt_conf.ddp_inodespace == 0)
943 /* don't enforce grant during recovery */
946 /* Update statfs data if required */
947 ofd_grant_statfs(env, exp, 1, NULL);
949 /* protect all grant counters */
950 spin_lock(&ofd->ofd_grant_lock);
952 /* fail precreate request if there is not enough blocks available for
954 if (ofd->ofd_osfs.os_bavail - (fed->fed_grant >> ofd->ofd_blockbits) <
955 (ofd->ofd_osfs.os_blocks >> 10)) {
956 spin_unlock(&ofd->ofd_grant_lock);
957 CDEBUG(D_RPCTRACE, "%s: not enough space for create "LPU64"\n",
958 ofd_obd(ofd)->obd_name,
959 ofd->ofd_osfs.os_bavail * ofd->ofd_osfs.os_blocks);
963 /* Grab free space from cached statfs data and take out space
964 * already granted to clients as well as reserved space */
965 left = ofd_grant_space_left(exp);
967 /* compute how much space is required to handle the precreation
969 wanted = *nr * ofd->ofd_dt_conf.ddp_inodespace;
970 if (wanted > fed->fed_grant + left) {
971 /* that's beyond what remains, adjust the number of objects that
972 * can be safely precreated */
973 wanted = fed->fed_grant + left;
974 *nr = wanted / ofd->ofd_dt_conf.ddp_inodespace;
976 /* we really have no space any more for precreation,
977 * fail the precreate request with ENOSPC */
978 spin_unlock(&ofd->ofd_grant_lock);
981 /* compute space needed for the new number of creations */
982 wanted = *nr * ofd->ofd_dt_conf.ddp_inodespace;
984 LASSERT(wanted <= fed->fed_grant + left);
986 if (wanted <= fed->fed_grant) {
987 /* we've enough grant space to handle this precreate request */
988 fed->fed_grant -= wanted;
990 /* we need to take some space from the ungranted pool */
991 ofd->ofd_tot_granted += wanted - fed->fed_grant;
992 left -= wanted - fed->fed_grant;
995 info->fti_used = wanted;
996 fed->fed_pending += info->fti_used;
997 ofd->ofd_tot_pending += info->fti_used;
999 /* grant more space for precreate purpose if possible. */
1000 wanted = OST_MAX_PRECREATE * ofd->ofd_dt_conf.ddp_inodespace / 2;
1001 if (wanted > fed->fed_grant) {
1002 /* always try to book enough space to handle a large precreate
1004 wanted -= fed->fed_grant;
1005 ofd_grant(exp, fed->fed_grant, wanted, left, false);
1007 spin_unlock(&ofd->ofd_grant_lock);
1012 * Called at commit time to update pending grant counter for writes in flight
1014 * \param env - is the lu environment provided by the caller
1015 * \param exp - is the export of the client which sent the request
1017 void ofd_grant_commit(const struct lu_env *env, struct obd_export *exp,
1020 struct ofd_device *ofd = ofd_exp(exp);
1021 struct ofd_thread_info *info = ofd_info(env);
1022 unsigned long pending;
1026 /* get space accounted in tot_pending for the I/O, set in
1027 * ofd_grant_check() */
1028 pending = info->fti_used;
1032 spin_lock(&ofd->ofd_grant_lock);
1033 /* Don't update statfs data for errors raised before commit (e.g.
1034 * bulk transfer failed, ...) since we know those writes have not been
1035 * processed. For other errors hit during commit, we cannot really tell
1036 * whether or not something was written, so we update statfs data.
1037 * In any case, this should not be fatal since we always get fresh
1038 * statfs data before failing a request with ENOSPC */
1040 spin_lock(&ofd->ofd_osfs_lock);
1041 /* Take pending out of cached statfs data */
1042 ofd->ofd_osfs.os_bavail -= min_t(obd_size,
1043 ofd->ofd_osfs.os_bavail,
1044 pending >> ofd->ofd_blockbits);
1045 if (ofd->ofd_statfs_inflight)
1046 /* someone is running statfs and want to be notified of
1047 * writes happening meanwhile */
1048 ofd->ofd_osfs_inflight += pending;
1049 spin_unlock(&ofd->ofd_osfs_lock);
1052 if (exp->exp_filter_data.fed_pending < pending) {
1053 CERROR("%s: cli %s/%p fed_pending(%lu) < grant_used(%lu)\n",
1054 exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
1055 exp->exp_filter_data.fed_pending, pending);
1056 spin_unlock(&ofd->ofd_grant_lock);
1059 exp->exp_filter_data.fed_pending -= pending;
1061 if (ofd->ofd_tot_granted < pending) {
1062 CERROR("%s: cli %s/%p tot_granted("LPU64") < grant_used(%lu)"
1063 "\n", exp->exp_obd->obd_name,
1064 exp->exp_client_uuid.uuid, exp, ofd->ofd_tot_granted,
1066 spin_unlock(&ofd->ofd_grant_lock);
1069 ofd->ofd_tot_granted -= pending;
1071 if (ofd->ofd_tot_pending < pending) {
1072 CERROR("%s: cli %s/%p tot_pending("LPU64") < grant_used(%lu)"
1073 "\n", exp->exp_obd->obd_name, exp->exp_client_uuid.uuid,
1074 exp, ofd->ofd_tot_pending, pending);
1075 spin_unlock(&ofd->ofd_grant_lock);
1078 ofd->ofd_tot_pending -= pending;
1079 spin_unlock(&ofd->ofd_grant_lock);