-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*/
/*
#define DEBUG_SUBSYSTEM S_FILTER
-#ifndef AUTOCONF_INCLUDED
-#include <linux/config.h>
-#endif
#include <linux/module.h>
#include <linux/pagemap.h> // XXX kill me soon
#include <linux/version.h>
fed->fed_grant -= oa->o_dropped;
fed->fed_dirty = oa->o_dirty;
- if (oa->o_flags & OBD_FL_SHRINK_GRANT) {
+ if (oa->o_valid & OBD_MD_FLFLAGS && oa->o_flags & OBD_FL_SHRINK_GRANT) {
obd_size left_space = filter_grant_space_left(exp);
struct filter_obd *filter = &exp->exp_obd->u.filter;
CERROR("%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
obd->obd_name, exp->exp_client_uuid.uuid, exp,
fed->fed_dirty, fed->fed_pending, fed->fed_grant);
- spin_unlock(&obd->obd_osfs_lock);
+ cfs_spin_unlock(&obd->obd_osfs_lock);
LBUG();
}
EXIT;
LASSERT_SPIN_LOCKED(&obd->obd_osfs_lock);
- if (cfs_time_before_64(obd->obd_osfs_age, cfs_time_current_64() - HZ)) {
+ if (cfs_time_before_64(obd->obd_osfs_age,
+ cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
restat:
rc = fsfilt_statfs(obd, obd->u.obt.obt_sb,
- cfs_time_current_64() + HZ);
+ cfs_time_shift_64(OBD_STATFS_CACHE_SECONDS));
if (rc) /* N.B. statfs can't really fail */
RETURN(0);
statfs_done = 1;
"current"LPU64"\n",
obd->obd_name, exp->exp_client_uuid.uuid,
exp, fed->fed_grant, want,current_grant);
- spin_unlock(&obd->obd_osfs_lock);
+ cfs_spin_unlock(&obd->obd_osfs_lock);
LBUG();
}
}
* as we might end up waiting on a page he sent in the request we're serving.
* use __GFP_HIGHMEM so that the pages can use all of the available memory
* on 32-bit machines
- * use more agressive GFP_HIGHUSER flags from non-local clients to be able to
- * generate more memory pressure, but at the same time use __GFP_NOMEMALLOC
- * in order not to exhaust emergency reserves.
+ * use more aggressive GFP_HIGHUSER flags from non-local clients to be able to
+ * generate more memory pressure.
*
* See Bug 19529 and Bug 19917 for details.
*/
LASSERT(inode != NULL);
for (i = 0; i < obj->ioo_bufcnt; i++, rnb++) {
-#ifdef HAVE_TRUNCATE_RANGE
/* remove pages in which range is fit */
truncate_inode_pages_range(inode->i_mapping,
rnb->offset & CFS_PAGE_MASK,
(rnb->offset + rnb->len - 1) |
~CFS_PAGE_MASK);
-#else
- /* use invalidate for old kernels */
- invalidate_mapping_pages(inode->i_mapping,
- rnb->offset >> CFS_PAGE_SHIFT,
- (rnb->offset + rnb->len) >>
- CFS_PAGE_SHIFT);
-#endif
}
}
LASSERTF(objcount == 1, "%d\n", objcount);
LASSERTF(obj->ioo_bufcnt > 0, "%d\n", obj->ioo_bufcnt);
- LASSERT(oa->o_valid & OBD_MD_FLGROUP);
- rc = filter_auth_capa(exp, NULL, oa->o_gr, capa,
+ rc = filter_auth_capa(exp, NULL, oa->o_seq, capa,
CAPA_OPC_OSS_READ);
if (rc)
RETURN(rc);
if (oa && oa->o_valid & OBD_MD_FLGRANT) {
- spin_lock(&obd->obd_osfs_lock);
+ cfs_spin_lock(&obd->obd_osfs_lock);
filter_grant_incoming(exp, oa);
- if (!(oa->o_flags & OBD_FL_SHRINK_GRANT))
+ if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
+ !(oa->o_flags & OBD_FL_SHRINK_GRANT))
oa->o_grant = 0;
- spin_unlock(&obd->obd_osfs_lock);
+ cfs_spin_unlock(&obd->obd_osfs_lock);
}
iobuf = filter_iobuf_get(&obd->u.filter, oti);
RETURN(PTR_ERR(iobuf));
push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- dentry = filter_oa2dentry(obd, oa);
+ dentry = filter_oa2dentry(obd, &oa->o_oi);
if (IS_ERR(dentry)) {
rc = PTR_ERR(dentry);
dentry = NULL;
fsfilt_check_slow(obd, now, "preprw_read setup");
/* find pages for all segments, fill array with them */
- do_gettimeofday(&start);
+ cfs_gettimeofday(&start);
for (i = 0, lnb = res; i < *npages; i++, lnb++) {
lnb->dentry = dentry;
lprocfs_counter_add(obd->obd_stats, LPROC_FILTER_CACHE_MISS, 1);
filter_iobuf_add_page(obd, iobuf, inode, lnb->page);
}
- do_gettimeofday(&end);
+ cfs_gettimeofday(&end);
timediff = cfs_timeval_sub(&end, &start, NULL);
lprocfs_counter_add(obd->obd_stats, LPROC_FILTER_GET_PAGE, timediff);
if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats)
lprocfs_counter_add(exp->exp_nid_stats->nid_stats,
LPROC_FILTER_READ_BYTES, tot_bytes);
-
+ filter_counter_incr(exp, LPROC_FILTER_STATS_READ,
+ oti ? oti->oti_jobid : NULL, tot_bytes);
EXIT;
cleanup:
int blocksize = exp->exp_obd->u.obt.obt_sb->s_blocksize;
unsigned long used = 0, ungranted = 0, using;
int i, rc = -ENOSPC, obj, n = 0;
+ int resend = 0;
+
+ if ((oa->o_valid & OBD_MD_FLFLAGS) &&
+ (oa->o_flags & OBD_FL_RECOV_RESEND)) {
+ resend = 1;
+ CDEBUG(D_CACHE, "Recoverable resend arrived, skipping "
+ "accounting\n");
+ }
LASSERT_SPIN_LOCKED(&exp->exp_obd->obd_osfs_lock);
if (tmp)
bytes += blocksize - tmp;
+ lnb[n].rc = 0;
if ((lnb[n].flags & OBD_BRW_FROM_GRANT) &&
(oa->o_valid & OBD_MD_FLGRANT)) {
- if (fed->fed_grant < used + bytes) {
+ if (resend) {
+ /* this is a recoverable resent */
+ lnb[n].flags |= OBD_BRW_GRANTED;
+ rc = 0;
+ continue;
+ } else if (fed->fed_grant < used + bytes) {
CDEBUG(D_CACHE,
"%s: cli %s/%p claims %ld+%d "
"GRANT, real grant %lu idx %d\n",
CERROR("%s: cli %s/%p dirty %ld pend %ld grant %ld\n",
exp->exp_obd->obd_name, exp->exp_client_uuid.uuid, exp,
fed->fed_dirty, fed->fed_pending, fed->fed_grant);
- spin_unlock(&exp->exp_obd->obd_osfs_lock);
+ cfs_spin_unlock(&exp->exp_obd->obd_osfs_lock);
LBUG();
}
return rc;
obd_size left;
unsigned long now = jiffies, timediff;
int rc = 0, i, tot_bytes = 0, cleanup_phase = 0, localreq = 0;
+ int retries = 0;
ENTRY;
LASSERT(objcount == 1);
LASSERT(obj->ioo_bufcnt > 0);
- LASSERT(oa->o_valid & OBD_MD_FLGROUP);
- rc = filter_auth_capa(exp, NULL, oa->o_gr, capa,
+ rc = filter_auth_capa(exp, NULL, oa->o_seq, capa,
CAPA_OPC_OSS_WRITE);
if (rc)
RETURN(rc);
GOTO(cleanup, rc = PTR_ERR(iobuf));
cleanup_phase = 1;
- dentry = filter_fid2dentry(obd, NULL, obj->ioo_gr,
+ dentry = filter_fid2dentry(obd, NULL, obj->ioo_seq,
obj->ioo_id);
if (IS_ERR(dentry))
GOTO(cleanup, rc = PTR_ERR(dentry));
cleanup_phase = 2;
if (dentry->d_inode == NULL) {
- CERROR("%s: trying to BRW to non-existent file "LPU64"\n",
- obd->obd_name, obj->ioo_id);
- GOTO(cleanup, rc = -ENOENT);
+ if (exp->exp_obd->obd_recovering) {
+ struct obdo *noa = oa;
+
+ if (oa == NULL) {
+ OBDO_ALLOC(noa);
+ if (noa == NULL)
+ GOTO(recreate_out, rc = -ENOMEM);
+ noa->o_id = obj->ioo_id;
+ noa->o_valid = OBD_MD_FLID;
+ }
+
+ if (filter_create(NULL, exp, noa, NULL, oti) == 0) {
+ f_dput(dentry);
+ dentry = filter_fid2dentry(exp->exp_obd, NULL,
+ obj->ioo_seq,
+ obj->ioo_id);
+ }
+ if (oa == NULL)
+ OBDO_FREE(noa);
+ }
+ recreate_out:
+ if (IS_ERR(dentry) || dentry->d_inode == NULL) {
+ CERROR("%s: BRW to missing obj "LPU64"/"LPU64":rc %d\n",
+ exp->exp_obd->obd_name,
+ obj->ioo_id, obj->ioo_seq,
+ IS_ERR(dentry) ? (int)PTR_ERR(dentry) : -ENOENT);
+ if (IS_ERR(dentry))
+ cleanup_phase = 1;
+ GOTO(cleanup, rc = -ENOENT);
+ }
}
if (oa->o_valid & (OBD_MD_FLUID | OBD_MD_FLGID) &&
dentry->d_inode->i_mode & (S_ISUID | S_ISGID)) {
- rc = filter_capa_fixoa(exp, oa, oa->o_gr, capa);
+ rc = filter_capa_fixoa(exp, oa, oa->o_seq, capa);
if (rc)
GOTO(cleanup, rc);
}
fsfilt_check_slow(obd, now, "preprw_write setup");
+ /* Filter truncate first locks i_mutex then partially truncated
+ * page, filter write code first locks pages then take
+ * i_mutex. To avoid a deadlock in case of concurrent
+ * punch/write requests from one client, filter writes and
+ * filter truncates are serialized by i_alloc_sem, allowing
+ * multiple writes or single truncate. */
+ down_read(&dentry->d_inode->i_alloc_sem);
+ fsfilt_check_slow(obd, now, "i_alloc_sem");
+
/* Don't update inode timestamps if this write is older than a
* setattr which modifies the timestamps. b=10150 */
/* XXX when we start having persistent reservations this needs to
* be changed to filter_fmd_get() to create the fmd if it doesn't
* already exist so we can store the reservation handle there. */
- fmd = filter_fmd_find(exp, obj->ioo_id, obj->ioo_gr);
-
- LASSERT(oa != NULL);
- spin_lock(&obd->obd_osfs_lock);
- filter_grant_incoming(exp, oa);
- if (fmd && fmd->fmd_mactime_xid > oti->oti_xid)
- oa->o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLCTIME |
- OBD_MD_FLATIME);
- else
- obdo_to_inode(dentry->d_inode, oa, OBD_MD_FLATIME |
- OBD_MD_FLMTIME | OBD_MD_FLCTIME);
- cleanup_phase = 3;
-
- left = filter_grant_space_left(exp);
-
- fso.fso_dentry = dentry;
- fso.fso_bufcnt = *npages;
-
- rc = filter_grant_check(exp, oa, objcount, &fso, *npages, res,
- &left, dentry->d_inode);
+ fmd = filter_fmd_find(exp, obj->ioo_id, obj->ioo_seq);
+
+ LASSERT(oa != NULL);
+retry:
+ cfs_spin_lock(&obd->obd_osfs_lock);
+ if (retries == 0)
+ filter_grant_incoming(exp, oa);
+ if (fmd && fmd->fmd_mactime_xid > oti->oti_xid)
+ oa->o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLCTIME |
+ OBD_MD_FLATIME);
+ else
+ obdo_to_inode(dentry->d_inode, oa, OBD_MD_FLATIME |
+ OBD_MD_FLMTIME | OBD_MD_FLCTIME);
+ cleanup_phase = 3;
+
+ left = filter_grant_space_left(exp);
+
+ fso.fso_dentry = dentry;
+ fso.fso_bufcnt = *npages;
+
+ rc = filter_grant_check(exp, oa, objcount, &fso, *npages, res,
+ &left, dentry->d_inode);
+
+ /* do not zero out oa->o_valid as it is used in filter_commitrw_write()
+ * for setting UID/GID and fid EA in first write time. */
+ /* If OBD_FL_SHRINK_GRANT is set, the client just returned us some grant
+ * so no sense in allocating it some more. We either return the grant
+ * back to the client if we have plenty of space or we don't return
+ * anything if we are short. This was decided in filter_grant_incoming*/
+ if ((retries == 0) && (oa->o_valid & OBD_MD_FLGRANT) &&
+ (!(oa->o_valid & OBD_MD_FLFLAGS) ||
+ !(oa->o_flags & OBD_FL_SHRINK_GRANT)))
+ oa->o_grant = filter_grant(exp, oa->o_grant, oa->o_undirty,
+ left, 1);
+
+ cfs_spin_unlock(&obd->obd_osfs_lock);
+
+ if (rc == -ENOSPC && retries == 0) {
+ void *handle = NULL;
+
+ CDEBUG(D_INODE, "retry after commit pending journals");
+
+ retries = 1;
+ handle = fsfilt_start(obd, dentry->d_inode,
+ FSFILT_OP_SETATTR, NULL);
+ if (handle != NULL &&
+ fsfilt_commit(obd, dentry->d_inode, handle, 1) == 0)
+ goto retry;
+ }
- /* do not zero out oa->o_valid as it is used in filter_commitrw_write()
- * for setting UID/GID and fid EA in first write time. */
- if (oa->o_valid & OBD_MD_FLGRANT)
- oa->o_grant = filter_grant(exp, oa->o_grant, oa->o_undirty,
- left, 1);
-
- spin_unlock(&obd->obd_osfs_lock);
filter_fmd_put(exp, fmd);
+ OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK2, (obd_timeout + 1) / 4);
+
if (rc)
GOTO(cleanup, rc);
cleanup_phase = 4;
- /* Filter truncate first locks i_mutex then partally truncated
- * page, filter write code first locks pages then take
- * i_mutex. To avoid a deadlock in case of concurrent
- * punch/write requests from one client, filter writes and
- * filter truncates are serialized by i_alloc_sem, allowing
- * multiple writes or single truncate. */
- down_read(&dentry->d_inode->i_alloc_sem);
-
- do_gettimeofday(&start);
+ cfs_gettimeofday(&start);
for (i = 0, lnb = res; i < *npages; i++, lnb++) {
/* We still set up for ungranted pages so that granted pages
if (lnb->rc == 0)
tot_bytes += lnb->len;
}
- do_gettimeofday(&end);
+ cfs_gettimeofday(&end);
timediff = cfs_timeval_sub(&end, &start, NULL);
lprocfs_counter_add(obd->obd_stats, LPROC_FILTER_GET_PAGE, timediff);
fsfilt_check_slow(obd, now, "start_page_write");
+ lprocfs_counter_add(obd->obd_stats, LPROC_FILTER_WRITE_BYTES,
+ tot_bytes);
+
if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats)
lprocfs_counter_add(exp->exp_nid_stats->nid_stats,
LPROC_FILTER_WRITE_BYTES, tot_bytes);
+ filter_counter_incr(exp, LPROC_FILTER_STATS_WRITE,
+ oti ? oti->oti_jobid : NULL, tot_bytes);
EXIT;
cleanup:
switch(cleanup_phase) {
lnb->page = NULL;
}
}
- up_read(&dentry->d_inode->i_alloc_sem);
}
case 3:
+ if (rc)
+ up_read(&dentry->d_inode->i_alloc_sem);
+
filter_iobuf_put(&obd->u.filter, iobuf, oti);
case 2:
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
case 1:
filter_iobuf_put(&obd->u.filter, iobuf, oti);
case 0:
- spin_lock(&obd->obd_osfs_lock);
+ cfs_spin_lock(&obd->obd_osfs_lock);
if (oa)
filter_grant_incoming(exp, oa);
- spin_unlock(&obd->obd_osfs_lock);
+ cfs_spin_unlock(&obd->obd_osfs_lock);
pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
break;
default:;
return rc;
}
-int filter_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
- int objcount, struct obd_ioobj *obj,
+int filter_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
+ struct obdo *oa, int objcount, struct obd_ioobj *obj,
struct niobuf_remote *nb, int *npages,
struct niobuf_local *res, struct obd_trans_info *oti,
struct lustre_capa *capa)
int i;
ENTRY;
- osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
+ osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id);
/* If oa != NULL then filter_preprw_read updated the inode atime
* and we should update the lvb so that other glimpses will also
* get the updated value. bug 5972 */
unsigned long pending = 0;
int i;
- spin_lock(&exp->exp_obd->obd_osfs_lock);
+ cfs_spin_lock(&exp->exp_obd->obd_osfs_lock);
for (i = 0, lnb = res; i < niocount; i++, lnb++)
pending += lnb->lnb_grant_used;
filter->fo_tot_pending, pending);
filter->fo_tot_pending -= pending;
- spin_unlock(&exp->exp_obd->obd_osfs_lock);
+ cfs_spin_unlock(&exp->exp_obd->obd_osfs_lock);
}
-int filter_commitrw(int cmd, struct obd_export *exp, struct obdo *oa,
- int objcount, struct obd_ioobj *obj,
+int filter_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
+ struct obdo *oa, int objcount, struct obd_ioobj *obj,
struct niobuf_remote *nb, int npages,
struct niobuf_local *res, struct obd_trans_info *oti,
int rc)
LBUG();
return -EPROTO;
}
-
-int filter_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
- obd_count oa_bufs, struct brw_page *pga,
- struct obd_trans_info *oti)
-{
- struct obd_ioobj ioo;
- struct niobuf_local *lnb;
- struct niobuf_remote *rnb;
- obd_count i;
- int ret = 0, npages;
- ENTRY;
-
- OBD_ALLOC(lnb, oa_bufs * sizeof(struct niobuf_local));
- OBD_ALLOC(rnb, oa_bufs * sizeof(struct niobuf_remote));
-
- if (lnb == NULL || rnb == NULL)
- GOTO(out, ret = -ENOMEM);
-
- for (i = 0; i < oa_bufs; i++) {
- lnb[i].page = pga[i].pg;
- rnb[i].offset = pga[i].off;
- rnb[i].len = pga[i].count;
- }
-
- obdo_to_ioobj(oinfo->oi_oa, &ioo);
- ioo.ioo_bufcnt = oa_bufs;
-
- npages = oa_bufs;
- ret = filter_preprw(cmd, exp, oinfo->oi_oa, 1, &ioo,
- rnb, &npages, lnb, oti, oinfo_capa(oinfo));
- if (ret != 0)
- GOTO(out, ret);
- LASSERTF(oa_bufs == npages, "%u != %u\n", oa_bufs, npages);
-
- ret = filter_commitrw(cmd, exp, oinfo->oi_oa, 1, &ioo, rnb,
- npages, lnb, oti, ret);
-
-out:
- if (lnb)
- OBD_FREE(lnb, oa_bufs * sizeof(struct niobuf_local));
- if (rnb)
- OBD_FREE(rnb, oa_bufs * sizeof(struct niobuf_remote));
- RETURN(ret);
-}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff