1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light block IO
6 * Copyright (c) 2002-2004 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
30 #include <sys/types.h>
32 #include <sys/queue.h>
49 #include "llite_lib.h"
53 struct obd_io_group *lig_oig;
54 struct inode *lig_inode;
55 struct lustre_rw_params *lig_params;
59 struct ll_async_page *lig_llaps;
60 struct page *lig_pages;
61 void *lig_llap_cookies;
64 #define LLU_IO_GROUP_SIZE(x) \
65 (sizeof(struct llu_io_group) + \
66 (sizeof(struct ll_async_page) + \
67 sizeof(struct page) + \
68 llap_cookie_size) * (x))
72 struct inode *lis_inode;
76 struct llu_io_group *lis_groups[0];
78 #define LLU_IO_SESSION_SIZE(x) \
79 (sizeof(struct llu_io_session) + (x) * 2 * sizeof(void *))
82 typedef ssize_t llu_file_piov_t(const struct iovec *iovec, int iovlen,
83 _SYSIO_OFF_T pos, ssize_t len,
86 size_t llap_cookie_size;
88 static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock)
90 struct llu_inode_info *lli = llu_i2info(inode);
91 struct lov_stripe_md *lsm = lli->lli_smd;
92 struct obd_export *exp = llu_i2obdexp(inode);
95 struct ldlm_lock *lock;
96 struct lov_stripe_md *lsm;
97 } key = { .name = "lock_to_stripe", .lock = lock, .lsm = lsm };
98 __u32 stripe, vallen = sizeof(stripe);
102 if (lsm->lsm_stripe_count == 1)
105 /* get our offset in the lov */
106 rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe);
108 CERROR("obd_get_info: rc = %d\n", rc);
111 LASSERT(stripe < lsm->lsm_stripe_count);
115 int llu_extent_lock_cancel_cb(struct ldlm_lock *lock,
116 struct ldlm_lock_desc *new, void *data,
119 struct lustre_handle lockh = { 0 };
123 if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
124 LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
129 case LDLM_CB_BLOCKING:
130 ldlm_lock2handle(lock, &lockh);
131 rc = ldlm_cli_cancel(&lockh);
133 CERROR("ldlm_cli_cancel failed: %d\n", rc);
135 case LDLM_CB_CANCELING: {
137 struct llu_inode_info *lli;
138 struct lov_stripe_md *lsm;
142 /* This lock wasn't granted, don't try to evict pages */
143 if (lock->l_req_mode != lock->l_granted_mode)
146 inode = llu_inode_from_lock(lock);
149 lli= llu_i2info(inode);
156 stripe = llu_lock_to_stripe_offset(inode, lock);
157 lock_res_and_lock(lock);
158 kms = ldlm_extent_shift_kms(lock,
159 lsm->lsm_oinfo[stripe]->loi_kms);
160 unlock_res_and_lock(lock);
161 if (lsm->lsm_oinfo[stripe]->loi_kms != kms)
162 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
163 lsm->lsm_oinfo[stripe]->loi_kms, kms);
164 lsm->lsm_oinfo[stripe]->loi_kms = kms;
176 static int llu_glimpse_callback(struct ldlm_lock *lock, void *reqp)
178 struct ptlrpc_request *req = reqp;
179 struct inode *inode = llu_inode_from_lock(lock);
180 struct llu_inode_info *lli;
186 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
187 lli = llu_i2info(inode);
189 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
190 if (lli->lli_smd == NULL)
191 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
193 /* First, find out which stripe index this lock corresponds to. */
194 if (lli->lli_smd->lsm_stripe_count > 1)
195 stripe = llu_lock_to_stripe_offset(inode, lock);
197 req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
198 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
200 rc = req_capsule_server_pack(&req->rq_pill);
202 CERROR("failed pack reply: %d\n", rc);
206 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
207 lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms;
209 LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64,
210 (__u64)llu_i2stat(inode)->st_size, stripe,lvb->lvb_size);
214 /* These errors are normal races, so we don't want to fill the console
215 * with messages by calling ptlrpc_error() */
216 if (rc == -ELDLM_NO_LOCK_DATA)
217 lustre_pack_reply(req, 1, NULL, NULL);
223 static int llu_merge_lvb(struct inode *inode)
225 struct llu_inode_info *lli = llu_i2info(inode);
226 struct llu_sb_info *sbi = llu_i2sbi(inode);
227 struct intnl_stat *st = llu_i2stat(inode);
232 inode_init_lvb(inode, &lvb);
233 rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
234 st->st_size = lvb.lvb_size;
235 st->st_blocks = lvb.lvb_blocks;
236 /* handle st_blocks overflow gracefully */
237 if (st->st_blocks < lvb.lvb_blocks)
238 st->st_blocks = ~0UL;
239 st->st_mtime = lvb.lvb_mtime;
240 st->st_atime = lvb.lvb_atime;
241 st->st_ctime = lvb.lvb_ctime;
246 int llu_local_size(struct inode *inode)
248 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
249 struct llu_inode_info *lli = llu_i2info(inode);
250 struct llu_sb_info *sbi = llu_i2sbi(inode);
251 struct lustre_handle lockh = { 0 };
256 if (lli->lli_smd->lsm_stripe_count == 0)
259 rc = obd_match(sbi->ll_dt_exp, lli->lli_smd, LDLM_EXTENT,
260 &policy, LCK_PR, &flags, inode, &lockh);
266 rc = llu_merge_lvb(inode);
267 obd_cancel(sbi->ll_dt_exp, lli->lli_smd, LCK_PR, &lockh);
271 /* NB: lov_merge_size will prefer locally cached writes if they extend the
272 * file (because it prefers KMS over RSS when larger) */
273 int llu_glimpse_size(struct inode *inode)
275 struct llu_inode_info *lli = llu_i2info(inode);
276 struct intnl_stat *st = llu_i2stat(inode);
277 struct llu_sb_info *sbi = llu_i2sbi(inode);
278 struct lustre_handle lockh = { 0 };
279 struct ldlm_enqueue_info einfo = { 0 };
280 struct obd_info oinfo = { { { 0 } } };
284 /* If size is cached on the mds, skip glimpse. */
285 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
288 CDEBUG(D_DLMTRACE, "Glimpsing inode "LPU64"\n", (__u64)st->st_ino);
291 CDEBUG(D_DLMTRACE, "No objects for inode "LPU64"\n",
296 einfo.ei_type = LDLM_EXTENT;
297 einfo.ei_mode = LCK_PR;
298 einfo.ei_cb_bl = osc_extent_blocking_cb;
299 einfo.ei_cb_cp = ldlm_completion_ast;
300 einfo.ei_cb_gl = llu_glimpse_callback;
301 einfo.ei_cbdata = inode;
303 oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF;
304 oinfo.oi_lockh = &lockh;
305 oinfo.oi_md = lli->lli_smd;
306 oinfo.oi_flags = LDLM_FL_HAS_INTENT;
308 rc = obd_enqueue_rqset(sbi->ll_dt_exp, &oinfo, &einfo);
310 CERROR("obd_enqueue returned rc %d, returning -EIO\n", rc);
311 RETURN(rc > 0 ? -EIO : rc);
314 rc = llu_merge_lvb(inode);
315 CDEBUG(D_DLMTRACE, "glimpse: size: "LPU64", blocks: "LPU64"\n",
316 (__u64)st->st_size, (__u64)st->st_blocks);
321 int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
322 struct lov_stripe_md *lsm, int mode,
323 ldlm_policy_data_t *policy, struct lustre_handle *lockh,
326 struct llu_sb_info *sbi = llu_i2sbi(inode);
327 struct intnl_stat *st = llu_i2stat(inode);
328 struct ldlm_enqueue_info einfo = { 0 };
329 struct obd_info oinfo = { { { 0 } } };
334 LASSERT(!lustre_handle_is_used(lockh));
335 CLASSERT(ELDLM_OK == 0);
337 /* XXX phil: can we do this? won't it screw the file size up? */
338 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
339 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
342 CDEBUG(D_DLMTRACE, "Locking inode %llu, start "LPU64" end "LPU64"\n",
343 (__u64)st->st_ino, policy->l_extent.start,
344 policy->l_extent.end);
346 einfo.ei_type = LDLM_EXTENT;
347 einfo.ei_mode = mode;
348 einfo.ei_cb_bl = osc_extent_blocking_cb;
349 einfo.ei_cb_cp = ldlm_completion_ast;
350 einfo.ei_cb_gl = llu_glimpse_callback;
351 einfo.ei_cbdata = inode;
353 oinfo.oi_policy = *policy;
354 oinfo.oi_lockh = lockh;
356 oinfo.oi_flags = ast_flags;
358 rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL);
359 *policy = oinfo.oi_policy;
363 inode_init_lvb(inode, &lvb);
364 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1);
365 if (policy->l_extent.start == 0 &&
366 policy->l_extent.end == OBD_OBJECT_EOF)
367 st->st_size = lvb.lvb_size;
370 st->st_mtime = lvb.lvb_mtime;
371 st->st_atime = lvb.lvb_atime;
372 st->st_ctime = lvb.lvb_ctime;
378 int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode,
379 struct lov_stripe_md *lsm, int mode,
380 struct lustre_handle *lockh)
382 struct llu_sb_info *sbi = llu_i2sbi(inode);
386 CLASSERT(ELDLM_OK == 0);
388 /* XXX phil: can we do this? won't it screw the file size up? */
389 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
390 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
393 rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh);
398 #define LLAP_MAGIC 12346789
400 struct ll_async_page {
404 struct page *llap_page;
405 struct inode *llap_inode;
408 static void llu_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
410 struct ll_async_page *llap;
412 struct lov_stripe_md *lsm;
413 obd_flag valid_flags;
416 llap = LLAP_FROM_COOKIE(data);
417 inode = llap->llap_inode;
418 lsm = llu_i2info(inode)->lli_smd;
420 oa->o_id = lsm->lsm_object_id;
421 oa->o_valid = OBD_MD_FLID;
422 valid_flags = OBD_MD_FLTYPE | OBD_MD_FLATIME;
423 if (cmd & OBD_BRW_WRITE)
424 valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
425 OBD_MD_FLUID | OBD_MD_FLGID |
426 OBD_MD_FLFID | OBD_MD_FLGENER;
428 obdo_from_inode(oa, inode, valid_flags);
432 static void llu_ap_update_obdo(void *data, int cmd, struct obdo *oa,
435 struct ll_async_page *llap;
438 llap = LLAP_FROM_COOKIE(data);
439 obdo_from_inode(oa, llap->llap_inode, valid);
444 /* called for each page in a completed rpc.*/
445 static int llu_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
447 struct ll_async_page *llap;
451 llap = LLAP_FROM_COOKIE(data);
452 llap->llap_queued = 0;
453 page = llap->llap_page;
456 if (cmd & OBD_BRW_WRITE)
457 CERROR("writeback error on page %p index %ld: %d\n",
458 page, page->index, rc);
463 static struct obd_capa * llu_ap_lookup_capa(void *data, int cmd)
468 static struct obd_async_page_ops llu_async_page_ops = {
469 .ap_make_ready = NULL,
470 .ap_refresh_count = NULL,
471 .ap_fill_obdo = llu_ap_fill_obdo,
472 .ap_update_obdo = llu_ap_update_obdo,
473 .ap_completion = llu_ap_completion,
474 .ap_lookup_capa = llu_ap_lookup_capa,
477 static int llu_queue_pio(int cmd, struct llu_io_group *group,
478 char *buf, size_t count, loff_t pos)
480 struct llu_inode_info *lli = llu_i2info(group->lig_inode);
481 struct intnl_stat *st = llu_i2stat(group->lig_inode);
482 struct lov_stripe_md *lsm = lli->lli_smd;
483 struct obd_export *exp = llu_i2obdexp(group->lig_inode);
484 struct page *pages = &group->lig_pages[group->lig_npages],*page = pages;
485 struct ll_async_page *llap = &group->lig_llaps[group->lig_npages];
486 void *llap_cookie = group->lig_llap_cookies +
487 llap_cookie_size * group->lig_npages;
488 int i, rc, npages = 0, ret_bytes = 0;
495 local_lock = group->lig_params->lrp_lock_mode != LCK_NL;
496 /* prepare the pages array */
498 unsigned long index, offset, bytes;
500 offset = (pos & ~CFS_PAGE_MASK);
501 index = pos >> CFS_PAGE_SHIFT;
502 bytes = CFS_PAGE_SIZE - offset;
506 /* prevent read beyond file range */
507 if (/* local_lock && */
508 cmd == OBD_BRW_READ && pos + bytes >= st->st_size) {
509 if (pos >= st->st_size)
511 bytes = st->st_size - pos;
514 /* prepare page for this index */
516 page->addr = buf - offset;
518 page->_offset = offset;
519 page->_count = bytes;
527 group->lig_rwcount += bytes;
531 group->lig_npages += npages;
533 for (i = 0, page = pages; i < npages;
534 i++, page++, llap++, llap_cookie += llap_cookie_size){
535 llap->llap_magic = LLAP_MAGIC;
536 llap->llap_cookie = llap_cookie;
537 rc = obd_prep_async_page(exp, lsm, NULL, page,
538 (obd_off)page->index << CFS_PAGE_SHIFT,
540 llap, &llap->llap_cookie,
541 1 /* no cache in liblustre at all */,
545 llap->llap_cookie = NULL;
549 CDEBUG(D_CACHE, "llap %p page %p group %p obj off "LPU64"\n",
550 llap, page, llap->llap_cookie,
551 (obd_off)pages->index << CFS_PAGE_SHIFT);
552 page->private = (unsigned long)llap;
553 llap->llap_page = page;
554 llap->llap_inode = group->lig_inode;
556 rc = obd_queue_group_io(exp, lsm, NULL, group->lig_oig,
557 llap->llap_cookie, cmd,
558 page->_offset, page->_count,
559 group->lig_params->lrp_brw_flags,
560 ASYNC_READY | ASYNC_URGENT |
561 ASYNC_COUNT_STABLE | ASYNC_GROUP_SYNC);
562 if (!local_lock && cmd == OBD_BRW_READ) {
564 * In OST-side locking case short reads cannot be
567 * The root of the problem is that
569 * kms = lov_merge_size(lsm, 1);
571 * glimpse_size(inode);
575 * logic in the read code (both llite and liblustre)
576 * only works correctly when client holds DLM lock on
577 * [start, end]. Without DLM lock KMS can be
578 * completely out of date, and client can either make
579 * spurious short-read (missing concurrent write), or
580 * return stale data (missing concurrent
581 * truncate). For llite client this is fatal, because
582 * incorrect data are cached and can be later sent
583 * back to the server (vide bug 5047). This is hard to
584 * fix by handling short-reads on the server, as there
585 * is no easy way to communicate file size (or amount
586 * of bytes read/written) back to the client,
587 * _especially_ because OSC pages can be sliced and
588 * dices into multiple RPCs arbitrary. Fortunately,
589 * liblustre doesn't cache data and the worst case is
590 * that we get race with concurrent write or truncate.
598 llap->llap_queued = 1;
605 struct llu_io_group * get_io_group(struct inode *inode, int maxpages,
606 struct lustre_rw_params *params)
608 struct llu_io_group *group;
611 if (!llap_cookie_size)
612 llap_cookie_size = obd_prep_async_page(llu_i2obdexp(inode),
617 OBD_ALLOC(group, LLU_IO_GROUP_SIZE(maxpages));
619 return ERR_PTR(-ENOMEM);
622 group->lig_inode = inode;
623 group->lig_maxpages = maxpages;
624 group->lig_params = params;
625 group->lig_llaps = (struct ll_async_page *)(group + 1);
626 group->lig_pages = (struct page *)(&group->lig_llaps[maxpages]);
627 group->lig_llap_cookies = (void *)(&group->lig_pages[maxpages]);
629 rc = oig_init(&group->lig_oig);
631 OBD_FREE(group, LLU_IO_GROUP_SIZE(maxpages));
638 static int max_io_pages(ssize_t len, int iovlen)
640 return (((len + CFS_PAGE_SIZE -1) / CFS_PAGE_SIZE) + 2 + iovlen - 1);
644 void put_io_group(struct llu_io_group *group)
646 struct lov_stripe_md *lsm = llu_i2info(group->lig_inode)->lli_smd;
647 struct obd_export *exp = llu_i2obdexp(group->lig_inode);
648 struct ll_async_page *llap = group->lig_llaps;
651 for (i = 0; i < group->lig_npages; i++, llap++) {
652 if (llap->llap_cookie)
653 obd_teardown_async_page(exp, lsm, NULL,
657 I_RELE(group->lig_inode);
659 oig_release(group->lig_oig);
660 OBD_FREE(group, LLU_IO_GROUP_SIZE(group->lig_maxpages));
664 ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen,
665 _SYSIO_OFF_T pos, ssize_t len,
668 struct llu_io_session *session = (struct llu_io_session *) private;
669 struct inode *inode = session->lis_inode;
670 struct llu_inode_info *lli = llu_i2info(inode);
671 struct intnl_stat *st = llu_i2stat(inode);
672 struct ll_file_data *fd = lli->lli_file_data;
673 struct lustre_handle lockh = {0};
674 struct lov_stripe_md *lsm = lli->lli_smd;
675 struct obd_export *exp = NULL;
676 struct llu_io_group *iogroup;
677 struct lustre_rw_params p;
680 int err, is_read, iovidx, ret;
682 ssize_t ret_len = len;
685 /* in a large iov read/write we'll be repeatedly called.
686 * so give a chance to answer cancel ast here
688 liblustre_wait_event(0);
690 exp = llu_i2obdexp(inode);
694 if (len == 0 || iovlen == 0)
697 if (pos + len > lli->lli_maxbytes)
700 lustre_build_lock_params(session->lis_cmd, lli->lli_open_flags,
701 lli->lli_sbi->ll_lco.lco_flags,
704 iogroup = get_io_group(inode, max_io_pages(len, iovlen), &p);
706 RETURN(PTR_ERR(iogroup));
708 local_lock = p.lrp_lock_mode != LCK_NL;
710 err = llu_extent_lock(fd, inode, lsm, p.lrp_lock_mode, &p.lrp_policy,
711 &lockh, p.lrp_ast_flags);
715 is_read = (session->lis_cmd == OBD_BRW_READ);
718 * If OST-side locking is used, KMS can be completely out of
719 * date, and, hence, cannot be used for short-read
720 * detection. Rely in OST to handle short reads in that case.
722 inode_init_lvb(inode, &lvb);
723 obd_merge_lvb(exp, lsm, &lvb, 1);
725 /* extent.end is last byte of the range */
726 if (p.lrp_policy.l_extent.end >= kms) {
727 /* A glimpse is necessary to determine whether
728 * we return a short read or some zeroes at
729 * the end of the buffer
731 * In the case of OST-side locking KMS can be
732 * completely out of date and short-reads maybe
733 * mishandled. See llu_queue_pio() for more detailed
736 if ((err = llu_glimpse_size(inode))) {
737 GOTO(err_unlock, err);
742 } else if (lli->lli_open_flags & O_APPEND) {
746 for (iovidx = 0; iovidx < iovlen; iovidx++) {
747 char *buf = (char *) iovec[iovidx].iov_base;
748 size_t count = iovec[iovidx].iov_len;
754 if (IS_BAD_PTR(buf) || IS_BAD_PTR(buf + count)) {
755 GOTO(err_unlock, err = -EFAULT);
759 if (/* local_lock && */ pos >= st->st_size)
762 if (pos >= lli->lli_maxbytes) {
763 GOTO(err_unlock, err = -EFBIG);
765 if (pos + count >= lli->lli_maxbytes)
766 count = lli->lli_maxbytes - pos;
769 ret = llu_queue_pio(session->lis_cmd, iogroup, buf, count, pos);
771 GOTO(err_unlock, err = ret);
775 LASSERT(ret == count);
776 obd_adjust_kms(exp, lsm, pos, 0);
777 /* file size grow immediately */
778 if (pos > st->st_size)
786 LASSERT(len == 0 || is_read); /* libsysio should guarantee this */
788 err = obd_trigger_group_io(exp, lsm, NULL, iogroup->lig_oig);
790 GOTO(err_unlock, err);
792 err = oig_wait(iogroup->lig_oig);
794 CERROR("%s error: %s\n", is_read ? "read" : "write", strerror(-err));
795 GOTO(err_unlock, err);
798 ret = llu_extent_unlock(fd, inode, lsm, p.lrp_lock_mode, &lockh);
800 CERROR("extent unlock error %d\n", ret);
802 session->lis_groups[session->lis_ngroups++] = iogroup;
806 llu_extent_unlock(fd, inode, lsm, p.lrp_lock_mode, &lockh);
808 put_io_group(iogroup);
809 RETURN((ssize_t)err);
813 struct llu_io_session *get_io_session(struct inode *ino, int ngroups, int cmd)
815 struct llu_io_session *session;
817 OBD_ALLOC(session, LLU_IO_SESSION_SIZE(ngroups));
822 session->lis_inode = ino;
823 session->lis_max_groups = ngroups;
824 session->lis_cmd = cmd;
828 static void put_io_session(struct llu_io_session *session)
832 for (i = 0; i < session->lis_ngroups; i++) {
833 if (session->lis_groups[i]) {
834 put_io_group(session->lis_groups[i]);
835 session->lis_groups[i] = NULL;
839 I_RELE(session->lis_inode);
840 OBD_FREE(session, LLU_IO_SESSION_SIZE(session->lis_max_groups));
843 static int llu_file_rwx(struct inode *ino,
847 struct llu_io_session *session;
849 int cmd = read ? OBD_BRW_READ : OBD_BRW_WRITE;
852 LASSERT(ioctx->ioctx_xtvlen >= 0);
853 LASSERT(ioctx->ioctx_iovlen >= 0);
855 liblustre_wait_event(0);
857 if (!ioctx->ioctx_xtvlen)
860 /* XXX consider other types later */
861 if (S_ISDIR(llu_i2stat(ino)->st_mode))
863 if (!S_ISREG(llu_i2stat(ino)->st_mode))
866 session = get_io_session(ino, ioctx->ioctx_xtvlen * 2, cmd);
870 cc = _sysio_enumerate_extents(ioctx->ioctx_xtv, ioctx->ioctx_xtvlen,
871 ioctx->ioctx_iov, ioctx->ioctx_iovlen,
872 llu_file_prwv, session);
875 LASSERT(!ioctx->ioctx_cc);
876 ioctx->ioctx_private = session;
879 put_io_session(session);
882 liblustre_wait_event(0);
886 int llu_iop_read(struct inode *ino,
890 struct intnl_stat *st = llu_i2stat(ino);
891 st->st_atime = CURRENT_TIME;
893 return llu_file_rwx(ino, ioctx, 1);
896 int llu_iop_write(struct inode *ino,
899 struct intnl_stat *st = llu_i2stat(ino);
900 st->st_mtime = st->st_ctime = CURRENT_TIME;
902 return llu_file_rwx(ino, ioctx, 0);
905 int llu_iop_iodone(struct ioctx *ioctx)
907 struct llu_io_session *session;
908 struct llu_io_group *group;
909 int i, err = 0, rc = 0;
912 liblustre_wait_event(0);
914 session = (struct llu_io_session *) ioctx->ioctx_private;
916 LASSERT(!IS_ERR(session));
918 for (i = 0; i < session->lis_ngroups; i++) {
919 group = session->lis_groups[i];
922 err = oig_wait(group->lig_oig);
927 ioctx->ioctx_cc += group->lig_rwcount;
929 session->lis_groups[i] = NULL;
935 ioctx->ioctx_cc = -1;
936 ioctx->ioctx_errno = -rc;
939 put_io_session(session);
940 ioctx->ioctx_private = NULL;
941 liblustre_wait_event(0);