1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdfilter/filter_io_26.c
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Andreas Dilger <adilger@clusterfs.com>
40 * Author: Phil Schwan <phil@clusterfs.com>
43 #ifndef AUTOCONF_INCLUDED
44 #include <linux/config.h>
46 #include <linux/module.h>
47 #include <linux/pagemap.h> // XXX kill me soon
48 #include <linux/version.h>
49 #include <linux/buffer_head.h>
51 #define DEBUG_SUBSYSTEM S_FILTER
53 #include <obd_class.h>
54 #include <lustre_fsfilt.h>
55 #include <lustre_quota.h>
56 #include "filter_internal.h"
58 /* 512byte block min */
59 #define MAX_BLOCKS_PER_PAGE (CFS_PAGE_SIZE / 512)
61 atomic_t dr_numreqs; /* number of reqs being processed */
62 wait_queue_head_t dr_wait;
66 struct page **dr_pages;
67 unsigned long *dr_blocks;
68 unsigned int dr_ignore_quota:1;
69 struct filter_obd *dr_filter;
72 static void record_start_io(struct filter_iobuf *iobuf, int rw, int size,
73 struct obd_export *exp)
75 struct filter_obd *filter = iobuf->dr_filter;
77 atomic_inc(&iobuf->dr_numreqs);
79 if (rw == OBD_BRW_READ) {
80 atomic_inc(&filter->fo_r_in_flight);
81 lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_RPC_HIST],
82 atomic_read(&filter->fo_r_in_flight));
83 lprocfs_oh_tally_log2(&filter->
84 fo_filter_stats.hist[BRW_R_DISK_IOSIZE],
86 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_brw_stats) {
87 lprocfs_oh_tally(&exp->exp_nid_stats->nid_brw_stats->
89 atomic_read(&filter->fo_r_in_flight));
90 lprocfs_oh_tally_log2(&exp->exp_nid_stats->
91 nid_brw_stats->hist[BRW_R_DISK_IOSIZE],
95 atomic_inc(&filter->fo_w_in_flight);
96 lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_W_RPC_HIST],
97 atomic_read(&filter->fo_w_in_flight));
98 lprocfs_oh_tally_log2(&filter->
99 fo_filter_stats.hist[BRW_W_DISK_IOSIZE],
101 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_brw_stats) {
102 lprocfs_oh_tally(&exp->exp_nid_stats->nid_brw_stats->
103 hist[BRW_W_RPC_HIST],
104 atomic_read(&filter->fo_r_in_flight));
105 lprocfs_oh_tally_log2(&exp->exp_nid_stats->
106 nid_brw_stats->hist[BRW_W_DISK_IOSIZE],
112 static void record_finish_io(struct filter_iobuf *iobuf, int rw, int rc)
114 struct filter_obd *filter = iobuf->dr_filter;
116 /* CAVEAT EMPTOR: possibly in IRQ context
117 * DO NOT record procfs stats here!!! */
119 if (rw == OBD_BRW_READ)
120 atomic_dec(&filter->fo_r_in_flight);
122 atomic_dec(&filter->fo_w_in_flight);
124 if (atomic_dec_and_test(&iobuf->dr_numreqs))
125 wake_up(&iobuf->dr_wait);
128 static int dio_complete_routine(struct bio *bio, unsigned int done, int error)
130 struct filter_iobuf *iobuf = bio->bi_private;
134 /* CAVEAT EMPTOR: possibly in IRQ context
135 * DO NOT record procfs stats here!!! */
137 if (bio->bi_size) /* Not complete */
140 if (unlikely(iobuf == NULL)) {
141 CERROR("***** bio->bi_private is NULL! This should never "
142 "happen. Normally, I would crash here, but instead I "
143 "will dump the bio contents to the console. Please "
144 "report this to <http://bugzilla.lustre.org/> , along "
145 "with any interesting messages leading up to this point "
146 "(like SCSI errors, perhaps). Because bi_private is "
147 "NULL, I can't wake up the thread that initiated this "
148 "IO - you will probably have to reboot this node.\n");
149 CERROR("bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d, "
150 "bi_idx: %d, bi->size: %d, bi_end_io: %p, bi_cnt: %d, "
151 "bi_private: %p\n", bio->bi_next, bio->bi_flags,
152 bio->bi_rw, bio->bi_vcnt, bio->bi_idx, bio->bi_size,
153 bio->bi_end_io, atomic_read(&bio->bi_cnt),
158 /* the check is outside of the cycle for performance reason -bzzz */
159 if (!test_bit(BIO_RW, &bio->bi_rw)) {
160 bio_for_each_segment(bvl, bio, i) {
161 if (likely(error == 0))
162 SetPageUptodate(bvl->bv_page);
163 LASSERT(PageLocked(bvl->bv_page));
164 ClearPageConstant(bvl->bv_page);
166 record_finish_io(iobuf, OBD_BRW_READ, error);
168 if (mapping_cap_page_constant_write(iobuf->dr_pages[0]->mapping)){
169 bio_for_each_segment(bvl, bio, i) {
170 ClearPageConstant(bvl->bv_page);
173 record_finish_io(iobuf, OBD_BRW_WRITE, error);
176 /* any real error is good enough -bzzz */
177 if (error != 0 && iobuf->dr_error == 0)
178 iobuf->dr_error = error;
180 /* Completed bios used to be chained off iobuf->dr_bios and freed in
181 * filter_clear_dreq(). It was then possible to exhaust the biovec-256
182 * mempool when serious on-disk fragmentation was encountered,
183 * deadlocking the OST. The bios are now released as soon as complete
184 * so the pool cannot be exhausted while IOs are competing. bug 10076 */
189 static int can_be_merged(struct bio *bio, sector_t sector)
196 size = bio->bi_size >> 9;
197 return bio->bi_sector + size == sector ? 1 : 0;
200 struct filter_iobuf *filter_alloc_iobuf(struct filter_obd *filter,
201 int rw, int num_pages)
203 struct filter_iobuf *iobuf;
205 LASSERTF(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ, "%x\n", rw);
207 OBD_ALLOC(iobuf, sizeof(*iobuf));
211 OBD_ALLOC(iobuf->dr_pages, num_pages * sizeof(*iobuf->dr_pages));
212 if (iobuf->dr_pages == NULL)
215 OBD_ALLOC(iobuf->dr_blocks,
216 MAX_BLOCKS_PER_PAGE * num_pages * sizeof(*iobuf->dr_blocks));
217 if (iobuf->dr_blocks == NULL)
220 iobuf->dr_filter = filter;
221 init_waitqueue_head(&iobuf->dr_wait);
222 atomic_set(&iobuf->dr_numreqs, 0);
223 iobuf->dr_max_pages = num_pages;
224 iobuf->dr_npages = 0;
230 OBD_FREE(iobuf->dr_pages,
231 num_pages * sizeof(*iobuf->dr_pages));
233 OBD_FREE(iobuf, sizeof(*iobuf));
235 RETURN(ERR_PTR(-ENOMEM));
238 static void filter_clear_iobuf(struct filter_iobuf *iobuf)
240 iobuf->dr_npages = 0;
242 atomic_set(&iobuf->dr_numreqs, 0);
245 void filter_free_iobuf(struct filter_iobuf *iobuf)
247 int num_pages = iobuf->dr_max_pages;
249 filter_clear_iobuf(iobuf);
251 OBD_FREE(iobuf->dr_blocks,
252 MAX_BLOCKS_PER_PAGE * num_pages * sizeof(*iobuf->dr_blocks));
253 OBD_FREE(iobuf->dr_pages,
254 num_pages * sizeof(*iobuf->dr_pages));
258 void filter_iobuf_put(struct filter_obd *filter, struct filter_iobuf *iobuf,
259 struct obd_trans_info *oti)
261 int thread_id = (oti && oti->oti_thread) ?
262 oti->oti_thread->t_id : -1;
264 if (unlikely(thread_id < 0)) {
265 filter_free_iobuf(iobuf);
269 LASSERTF(filter->fo_iobuf_pool[thread_id] == iobuf,
270 "iobuf mismatch for thread %d: pool %p iobuf %p\n",
271 thread_id, filter->fo_iobuf_pool[thread_id], iobuf);
272 filter_clear_iobuf(iobuf);
275 int filter_iobuf_add_page(struct obd_device *obd, struct filter_iobuf *iobuf,
276 struct inode *inode, struct page *page)
278 LASSERT(iobuf->dr_npages < iobuf->dr_max_pages);
279 iobuf->dr_pages[iobuf->dr_npages++] = page;
284 int filter_do_bio(struct obd_export *exp, struct inode *inode,
285 struct filter_iobuf *iobuf, int rw)
287 struct obd_device *obd = exp->exp_obd;
288 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
289 struct page **pages = iobuf->dr_pages;
290 int npages = iobuf->dr_npages;
291 unsigned long *blocks = iobuf->dr_blocks;
292 int total_blocks = npages * blocks_per_page;
293 int sector_bits = inode->i_sb->s_blocksize_bits - 9;
294 unsigned int blocksize = inode->i_sb->s_blocksize;
295 struct bio *bio = NULL;
297 unsigned long start_time = jiffies;
299 unsigned int page_offset;
308 LASSERT(iobuf->dr_npages == npages);
309 LASSERT(total_blocks <= OBDFILTER_CREATED_SCRATCHPAD_ENTRIES);
311 for (page_idx = 0, block_idx = 0;
313 page_idx++, block_idx += blocks_per_page) {
315 page = pages[page_idx];
316 LASSERT (block_idx + blocks_per_page <= total_blocks);
318 for (i = 0, page_offset = 0;
320 i += nblocks, page_offset += blocksize * nblocks) {
324 if (blocks[block_idx + i] == 0) { /* hole */
325 LASSERT(rw == OBD_BRW_READ);
326 memset(kmap(page) + page_offset, 0, blocksize);
331 sector = (sector_t)blocks[block_idx + i] << sector_bits;
333 /* Additional contiguous file blocks? */
334 while (i + nblocks < blocks_per_page &&
335 (sector + (nblocks << sector_bits)) ==
336 ((sector_t)blocks[block_idx + i + nblocks] <<
340 /* I only set the page to be constant only if it
341 * is mapped to a contiguous underlying disk block(s).
342 * It will then make sure the corresponding device
343 * cache of raid5 will be overwritten by this page.
345 if ((rw == OBD_BRW_WRITE) &&
346 (nblocks == blocks_per_page) &&
347 mapping_cap_page_constant_write(inode->i_mapping))
348 SetPageConstant(page);
351 can_be_merged(bio, sector) &&
352 bio_add_page(bio, page,
353 blocksize * nblocks, page_offset) != 0)
354 continue; /* added this frag OK */
358 bdev_get_queue(bio->bi_bdev);
360 /* Dang! I have to fragment this I/O */
361 CDEBUG(D_INODE, "bio++ sz %d vcnt %d(%d) "
362 "sectors %d(%d) psg %d(%d) hsg %d(%d)\n",
364 bio->bi_vcnt, bio->bi_max_vecs,
365 bio->bi_size >> 9, q->max_sectors,
366 bio_phys_segments(q, bio),
367 q->max_phys_segments,
368 bio_hw_segments(q, bio),
371 record_start_io(iobuf, rw, bio->bi_size, exp);
372 rc = fsfilt_send_bio(rw, obd, inode, bio);
374 CERROR("Can't send bio: %d\n", rc);
375 record_finish_io(iobuf, rw, rc);
381 /* allocate new bio, limited by max BIO size, b=9945 */
382 bio = bio_alloc(GFP_NOIO, max(BIO_MAX_PAGES,
383 (npages - page_idx) *
386 CERROR("Can't allocate bio %u*%u = %u pages\n",
387 (npages - page_idx), blocks_per_page,
388 (npages - page_idx) * blocks_per_page);
393 bio->bi_bdev = inode->i_sb->s_bdev;
394 bio->bi_sector = sector;
395 bio->bi_end_io = dio_complete_routine;
396 bio->bi_private = iobuf;
398 rc = bio_add_page(bio, page,
399 blocksize * nblocks, page_offset);
405 record_start_io(iobuf, rw, bio->bi_size, exp);
406 rc = fsfilt_send_bio(rw, obd, inode, bio);
411 CERROR("Can't send bio: %d\n", rc);
412 record_finish_io(iobuf, rw, rc);
417 wait_event(iobuf->dr_wait, atomic_read(&iobuf->dr_numreqs) == 0);
419 if (rw == OBD_BRW_READ) {
420 lprocfs_oh_tally(&obd->u.filter.fo_filter_stats.
421 hist[BRW_R_DIO_FRAGS],
423 lprocfs_oh_tally_log2(&obd->u.filter.
424 fo_filter_stats.hist[BRW_R_IO_TIME],
425 jiffies - start_time);
426 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_brw_stats) {
427 lprocfs_oh_tally(&exp->exp_nid_stats->nid_brw_stats->
428 hist[BRW_R_DIO_FRAGS],
430 lprocfs_oh_tally_log2(&exp->exp_nid_stats->
431 nid_brw_stats->hist[BRW_R_IO_TIME],
432 jiffies - start_time);
435 lprocfs_oh_tally(&obd->u.filter.fo_filter_stats.
436 hist[BRW_W_DIO_FRAGS], frags);
437 lprocfs_oh_tally_log2(&obd->u.filter.fo_filter_stats.
439 jiffies - start_time);
440 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_brw_stats) {
441 lprocfs_oh_tally(&exp->exp_nid_stats->nid_brw_stats->
442 hist[BRW_W_DIO_FRAGS],
444 lprocfs_oh_tally_log2(&exp->exp_nid_stats->
445 nid_brw_stats->hist[BRW_W_IO_TIME],
446 jiffies - start_time);
451 rc = iobuf->dr_error;
455 /* Must be called with i_mutex taken for writes; this will drop it */
456 int filter_direct_io(int rw, struct dentry *dchild, struct filter_iobuf *iobuf,
457 struct obd_export *exp, struct iattr *attr,
458 struct obd_trans_info *oti, void **wait_handle)
460 struct obd_device *obd = exp->exp_obd;
461 struct inode *inode = dchild->d_inode;
462 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
464 struct semaphore *sem;
467 LASSERTF(iobuf->dr_npages <= iobuf->dr_max_pages, "%d,%d\n",
468 iobuf->dr_npages, iobuf->dr_max_pages);
469 LASSERT(iobuf->dr_npages <= OBDFILTER_CREATED_SCRATCHPAD_ENTRIES);
471 if (rw == OBD_BRW_READ) {
472 if (iobuf->dr_npages == 0)
477 LASSERTF(rw == OBD_BRW_WRITE, "%x\n", rw);
478 LASSERT(iobuf->dr_npages > 0);
480 sem = &obd->u.filter.fo_alloc_lock;
482 lquota_enforce(filter_quota_interface_ref, obd,
483 iobuf->dr_ignore_quota);
486 rc = fsfilt_map_inode_pages(obd, inode, iobuf->dr_pages,
487 iobuf->dr_npages, iobuf->dr_blocks,
488 obdfilter_created_scratchpad, create, sem);
490 if (rw == OBD_BRW_WRITE) {
492 filter_tally(exp, iobuf->dr_pages,
493 iobuf->dr_npages, iobuf->dr_blocks,
495 if (attr->ia_size > i_size_read(inode))
496 attr->ia_valid |= ATTR_SIZE;
497 rc = fsfilt_setattr(obd, dchild,
498 oti->oti_handle, attr, 0);
501 UNLOCK_INODE_MUTEX(inode);
503 rc2 = filter_finish_transno(exp, inode, oti, 0, 0);
505 CERROR("can't close transaction: %d\n", rc2);
510 rc2 = fsfilt_commit_async(obd,inode,oti->oti_handle,
516 } else if (rc == 0) {
517 filter_tally(exp, iobuf->dr_pages, iobuf->dr_npages,
518 iobuf->dr_blocks, blocks_per_page, 0);
521 RETURN(filter_do_bio(exp, inode, iobuf, rw));
524 /* See if there are unallocated parts in given file region */
525 static int filter_range_is_mapped(struct inode *inode, obd_size offset, int len)
527 sector_t (*fs_bmap)(struct address_space *, sector_t) =
528 inode->i_mapping->a_ops->bmap;
531 /* We can't know if we are overwriting or not */
535 offset >>= inode->i_blkbits;
536 len >>= inode->i_blkbits;
538 for (j = 0; j <= len; j++)
539 if (fs_bmap(inode->i_mapping, offset + j) == 0)
546 * interesting use cases on how it interacts with VM:
548 * - vm writeout -- shouldn't see our pages as we don't mark them dirty
549 * though vm can find partial page left dirty by truncate. in this
550 * usual writeout is used unless our write rewrite that page - then we
551 * drop PG_dirty with PG_lock held.
556 int filter_commitrw_write(struct obd_export *exp, struct obdo *oa,
557 int objcount, struct obd_ioobj *obj,
558 struct niobuf_remote *nb, int niocount,
559 struct niobuf_local *res, struct obd_trans_info *oti,
562 struct niobuf_local *lnb;
563 struct filter_iobuf *iobuf = NULL;
564 struct lvfs_run_ctxt saved;
565 struct fsfilt_objinfo fso;
566 struct iattr iattr = { 0 };
567 struct inode *inode = res->dentry->d_inode;
568 unsigned long now = jiffies;
569 int i, err, cleanup_phase = 0;
570 struct obd_device *obd = exp->exp_obd;
571 struct filter_obd *fo = &obd->u.filter;
574 unsigned int qcids[MAXQUOTAS] = { oa->o_uid, oa->o_gid };
575 int rec_pending[MAXQUOTAS] = { 0, 0 }, quota_pages = 0;
578 LASSERT(oti != NULL);
579 LASSERT(objcount == 1);
580 LASSERT(current->journal_info == NULL);
585 iobuf = filter_iobuf_get(&obd->u.filter, oti);
587 GOTO(cleanup, rc = PTR_ERR(iobuf));
590 fso.fso_dentry = res->dentry;
591 fso.fso_bufcnt = obj->ioo_bufcnt;
593 iobuf->dr_ignore_quota = 0;
594 for (i = 0, lnb = res; i < niocount; i++, lnb++) {
596 __u32 flags = lnb->flags;
598 if (filter_range_is_mapped(inode, lnb->offset, lnb->len)) {
599 /* If overwriting an existing block,
600 * we don't need a grant */
601 if (!(flags & OBD_BRW_GRANTED) && lnb->rc == -ENOSPC)
607 if (lnb->rc) { /* ENOSPC, network RPC error, etc. */
608 CDEBUG(D_INODE, "Skipping [%d] == %d\n", i, lnb->rc);
612 LASSERT(PageLocked(lnb->page));
613 LASSERT(!PageWriteback(lnb->page));
615 /* preceding filemap_write_and_wait() should have clean pages */
616 if (fo->fo_writethrough_cache)
617 clear_page_dirty_for_io(lnb->page);
618 LASSERT(!PageDirty(lnb->page));
620 SetPageUptodate(lnb->page);
622 err = filter_iobuf_add_page(obd, iobuf, inode, lnb->page);
625 total_size += lnb->len;
627 /* we expect these pages to be in offset order, but we'll
629 this_size = lnb->offset + lnb->len;
630 if (this_size > iattr.ia_size)
631 iattr.ia_size = this_size;
633 /* if one page is a write-back page from client cache and
634 * not from direct_io, or it's written by root, then mark
635 * the whole io request as ignore quota request, remote
636 * client can not break through quota. */
637 if (exp_connect_rmtclient(exp))
638 flags &= ~OBD_BRW_NOQUOTA;
639 if ((flags & OBD_BRW_NOQUOTA) ||
640 (flags & (OBD_BRW_FROM_GRANT | OBD_BRW_SYNC)) ==
642 iobuf->dr_ignore_quota = 1;
645 /* we try to get enough quota to write here, and let ldiskfs
646 * decide if it is out of quota or not b=14783 */
647 lquota_chkquota(filter_quota_interface_ref, obd, qcids, rec_pending,
648 quota_pages, oti, LQUOTA_FLAGS_BLK, (void *)inode,
652 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
657 LOCK_INODE_MUTEX(inode);
658 fsfilt_check_slow(obd, now, "i_mutex");
659 oti->oti_handle = fsfilt_brw_start(obd, objcount, &fso, niocount, res,
661 if (IS_ERR(oti->oti_handle)) {
662 UNLOCK_INODE_MUTEX(inode);
663 rc = PTR_ERR(oti->oti_handle);
664 CDEBUG(rc == -ENOSPC ? D_INODE : D_ERROR,
665 "error starting transaction: rc = %d\n", rc);
666 oti->oti_handle = NULL;
669 /* have to call fsfilt_commit() from this point on */
671 fsfilt_check_slow(obd, now, "brw_start");
673 i = OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME;
675 /* If the inode still has SUID+SGID bits set (see filter_precreate())
676 * then we will accept the UID+GID if sent by the client for
677 * initializing the ownership of this inode. We only allow this to
678 * happen once (so clear these bits) and later only allow setattr. */
679 if (inode->i_mode & S_ISUID)
681 if (inode->i_mode & S_ISGID)
684 iattr_from_obdo(&iattr, oa, i);
685 if (iattr.ia_valid & (ATTR_UID | ATTR_GID)) {
688 CDEBUG(D_INODE, "update UID/GID to %lu/%lu\n",
689 (unsigned long)oa->o_uid, (unsigned long)oa->o_gid);
691 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
693 iattr.ia_valid |= ATTR_MODE;
694 iattr.ia_mode = inode->i_mode;
695 if (iattr.ia_valid & ATTR_UID)
696 iattr.ia_mode &= ~S_ISUID;
697 if (iattr.ia_valid & ATTR_GID)
698 iattr.ia_mode &= ~S_ISGID;
700 rc = filter_update_fidea(exp, inode, oti->oti_handle, oa);
702 /* To avoid problems with quotas, UID and GID must be set
703 * in the inode before filter_direct_io() - see bug 10357. */
704 save = iattr.ia_valid;
705 iattr.ia_valid &= (ATTR_UID | ATTR_GID);
706 rc = fsfilt_setattr(obd, res->dentry, oti->oti_handle,&iattr,0);
707 CDEBUG(D_QUOTA, "set uid(%u)/gid(%u) to ino(%lu). rc(%d)\n",
708 iattr.ia_uid, iattr.ia_gid, inode->i_ino, rc);
709 iattr.ia_valid = save & ~(ATTR_UID | ATTR_GID);
712 /* filter_direct_io drops i_mutex */
713 rc = filter_direct_io(OBD_BRW_WRITE, res->dentry, iobuf, exp, &iattr,
716 obdo_from_inode(oa, inode,
717 FILTER_VALID_FLAGS |OBD_MD_FLUID |OBD_MD_FLGID);
719 obdo_from_inode(oa, inode, OBD_MD_FLUID | OBD_MD_FLGID);
721 lquota_getflag(filter_quota_interface_ref, obd, oa);
723 fsfilt_check_slow(obd, now, "direct_io");
725 err = fsfilt_commit_wait(obd, inode, wait_handle);
727 CERROR("Failure to commit OST transaction (%d)?\n", err);
731 if (obd->obd_replayable && !rc)
732 LASSERTF(oti->oti_transno <= obd->obd_last_committed,
733 "oti_transno "LPU64" last_committed "LPU64"\n",
734 oti->oti_transno, obd->obd_last_committed);
736 fsfilt_check_slow(obd, now, "commitrw commit");
739 lquota_pending_commit(filter_quota_interface_ref, obd, qcids,
742 filter_grant_commit(exp, niocount, res);
744 switch (cleanup_phase) {
746 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
747 LASSERT(current->journal_info == NULL);
749 filter_iobuf_put(&obd->u.filter, iobuf, oti);
752 * lnb->page automatically returns back into per-thread page
758 /* trigger quota pre-acquire */
759 qcids[USRQUOTA] = oa->o_uid;
760 qcids[GRPQUOTA] = oa->o_gid;
761 err = lquota_adjust(filter_quota_interface_ref, obd, qcids, NULL, rc,
763 CDEBUG(err ? D_ERROR : D_QUOTA,
764 "filter adjust qunit! (rc:%d)\n", err);
766 for (i = 0, lnb = res; i < niocount; i++, lnb++) {
767 if (lnb->page == NULL)
770 LASSERT(PageLocked(lnb->page));
771 unlock_page(lnb->page);
773 page_cache_release(lnb->page);
777 if (inode && (fo->fo_writethrough_cache == 0 ||
778 i_size_read(inode) > fo->fo_readcache_max_filesize))
779 filter_invalidate_cache(obd, obj, nb, inode);
781 up_read(&inode->i_alloc_sem);