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 static int llu_extent_lock_callback(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;
182 int size[2] = { sizeof(struct ptlrpc_body), sizeof(*lvb) };
187 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
188 lli = llu_i2info(inode);
190 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
191 if (lli->lli_smd == NULL)
192 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
194 /* First, find out which stripe index this lock corresponds to. */
195 if (lli->lli_smd->lsm_stripe_count > 1)
196 stripe = llu_lock_to_stripe_offset(inode, lock);
198 rc = lustre_pack_reply(req, 2, size, NULL);
202 lvb = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*lvb));
203 lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms;
205 LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64,
206 (__u64)llu_i2stat(inode)->st_size, stripe,lvb->lvb_size);
210 /* These errors are normal races, so we don't want to fill the console
211 * with messages by calling ptlrpc_error() */
212 if (rc == -ELDLM_NO_LOCK_DATA)
213 lustre_pack_reply(req, 1, NULL, NULL);
219 static int llu_merge_lvb(struct inode *inode)
221 struct llu_inode_info *lli = llu_i2info(inode);
222 struct llu_sb_info *sbi = llu_i2sbi(inode);
223 struct intnl_stat *st = llu_i2stat(inode);
228 inode_init_lvb(inode, &lvb);
229 rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
230 st->st_size = lvb.lvb_size;
231 st->st_blocks = lvb.lvb_blocks;
232 /* handle st_blocks overflow gracefully */
233 if (st->st_blocks < lvb.lvb_blocks)
234 st->st_blocks = ~0UL;
235 st->st_mtime = lvb.lvb_mtime;
236 st->st_atime = lvb.lvb_atime;
237 st->st_ctime = lvb.lvb_ctime;
242 int llu_local_size(struct inode *inode)
244 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
245 struct llu_inode_info *lli = llu_i2info(inode);
246 struct llu_sb_info *sbi = llu_i2sbi(inode);
247 struct lustre_handle lockh = { 0 };
252 if (lli->lli_smd->lsm_stripe_count == 0)
255 rc = obd_match(sbi->ll_dt_exp, lli->lli_smd, LDLM_EXTENT,
256 &policy, LCK_PR, &flags, inode, &lockh);
262 rc = llu_merge_lvb(inode);
263 obd_cancel(sbi->ll_dt_exp, lli->lli_smd, LCK_PR, &lockh);
267 /* NB: lov_merge_size will prefer locally cached writes if they extend the
268 * file (because it prefers KMS over RSS when larger) */
269 int llu_glimpse_size(struct inode *inode)
271 struct llu_inode_info *lli = llu_i2info(inode);
272 struct intnl_stat *st = llu_i2stat(inode);
273 struct llu_sb_info *sbi = llu_i2sbi(inode);
274 struct lustre_handle lockh = { 0 };
275 struct ldlm_enqueue_info einfo = { 0 };
276 struct obd_info oinfo = { { { 0 } } };
280 /* If size is cached on the mds, skip glimpse. */
281 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
284 CDEBUG(D_DLMTRACE, "Glimpsing inode "LPU64"\n", (__u64)st->st_ino);
287 CDEBUG(D_DLMTRACE, "No objects for inode "LPU64"\n",
292 einfo.ei_type = LDLM_EXTENT;
293 einfo.ei_mode = LCK_PR;
294 einfo.ei_cb_bl = llu_extent_lock_callback;
295 einfo.ei_cb_cp = ldlm_completion_ast;
296 einfo.ei_cb_gl = llu_glimpse_callback;
297 einfo.ei_cbdata = inode;
299 oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF;
300 oinfo.oi_lockh = &lockh;
301 oinfo.oi_md = lli->lli_smd;
302 oinfo.oi_flags = LDLM_FL_HAS_INTENT;
304 rc = obd_enqueue_rqset(sbi->ll_dt_exp, &oinfo, &einfo);
306 CERROR("obd_enqueue returned rc %d, returning -EIO\n", rc);
307 RETURN(rc > 0 ? -EIO : rc);
310 rc = llu_merge_lvb(inode);
311 CDEBUG(D_DLMTRACE, "glimpse: size: "LPU64", blocks: "LPU64"\n",
312 (__u64)st->st_size, (__u64)st->st_blocks);
317 int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
318 struct lov_stripe_md *lsm, int mode,
319 ldlm_policy_data_t *policy, struct lustre_handle *lockh,
322 struct llu_sb_info *sbi = llu_i2sbi(inode);
323 struct intnl_stat *st = llu_i2stat(inode);
324 struct ldlm_enqueue_info einfo = { 0 };
325 struct obd_info oinfo = { { { 0 } } };
330 LASSERT(!lustre_handle_is_used(lockh));
331 CLASSERT(ELDLM_OK == 0);
333 /* XXX phil: can we do this? won't it screw the file size up? */
334 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
335 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
338 CDEBUG(D_DLMTRACE, "Locking inode %llu, start "LPU64" end "LPU64"\n",
339 (__u64)st->st_ino, policy->l_extent.start,
340 policy->l_extent.end);
342 einfo.ei_type = LDLM_EXTENT;
343 einfo.ei_mode = mode;
344 einfo.ei_cb_bl = llu_extent_lock_callback;
345 einfo.ei_cb_cp = ldlm_completion_ast;
346 einfo.ei_cb_gl = llu_glimpse_callback;
347 einfo.ei_cbdata = inode;
349 oinfo.oi_policy = *policy;
350 oinfo.oi_lockh = lockh;
352 oinfo.oi_flags = ast_flags;
354 rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL);
355 *policy = oinfo.oi_policy;
359 inode_init_lvb(inode, &lvb);
360 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1);
361 if (policy->l_extent.start == 0 &&
362 policy->l_extent.end == OBD_OBJECT_EOF)
363 st->st_size = lvb.lvb_size;
366 st->st_mtime = lvb.lvb_mtime;
367 st->st_atime = lvb.lvb_atime;
368 st->st_ctime = lvb.lvb_ctime;
374 int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode,
375 struct lov_stripe_md *lsm, int mode,
376 struct lustre_handle *lockh)
378 struct llu_sb_info *sbi = llu_i2sbi(inode);
382 CLASSERT(ELDLM_OK == 0);
384 /* XXX phil: can we do this? won't it screw the file size up? */
385 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
386 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
389 rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh);
394 #define LLAP_MAGIC 12346789
396 struct ll_async_page {
400 struct page *llap_page;
401 struct inode *llap_inode;
404 static void llu_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
406 struct ll_async_page *llap;
408 struct lov_stripe_md *lsm;
409 obd_flag valid_flags;
412 llap = LLAP_FROM_COOKIE(data);
413 inode = llap->llap_inode;
414 lsm = llu_i2info(inode)->lli_smd;
416 oa->o_id = lsm->lsm_object_id;
417 oa->o_valid = OBD_MD_FLID;
418 valid_flags = OBD_MD_FLTYPE | OBD_MD_FLATIME;
419 if (cmd & OBD_BRW_WRITE)
420 valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
421 OBD_MD_FLUID | OBD_MD_FLGID |
422 OBD_MD_FLFID | OBD_MD_FLGENER;
424 obdo_from_inode(oa, inode, valid_flags);
428 static void llu_ap_update_obdo(void *data, int cmd, struct obdo *oa,
431 struct ll_async_page *llap;
434 llap = LLAP_FROM_COOKIE(data);
435 obdo_from_inode(oa, llap->llap_inode, valid);
440 /* called for each page in a completed rpc.*/
441 static int llu_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
443 struct ll_async_page *llap;
447 llap = LLAP_FROM_COOKIE(data);
448 llap->llap_queued = 0;
449 page = llap->llap_page;
452 if (cmd & OBD_BRW_WRITE)
453 CERROR("writeback error on page %p index %ld: %d\n",
454 page, page->index, rc);
459 static struct obd_capa * llu_ap_lookup_capa(void *data, int cmd)
464 static struct obd_async_page_ops llu_async_page_ops = {
465 .ap_make_ready = NULL,
466 .ap_refresh_count = NULL,
467 .ap_fill_obdo = llu_ap_fill_obdo,
468 .ap_update_obdo = llu_ap_update_obdo,
469 .ap_completion = llu_ap_completion,
470 .ap_lookup_capa = llu_ap_lookup_capa,
473 static int llu_queue_pio(int cmd, struct llu_io_group *group,
474 char *buf, size_t count, loff_t pos)
476 struct llu_inode_info *lli = llu_i2info(group->lig_inode);
477 struct intnl_stat *st = llu_i2stat(group->lig_inode);
478 struct lov_stripe_md *lsm = lli->lli_smd;
479 struct obd_export *exp = llu_i2obdexp(group->lig_inode);
480 struct page *pages = &group->lig_pages[group->lig_npages],*page = pages;
481 struct ll_async_page *llap = &group->lig_llaps[group->lig_npages];
482 void *llap_cookie = group->lig_llap_cookies +
483 llap_cookie_size * group->lig_npages;
484 int i, rc, npages = 0, ret_bytes = 0;
491 local_lock = group->lig_params->lrp_lock_mode != LCK_NL;
492 /* prepare the pages array */
494 unsigned long index, offset, bytes;
496 offset = (pos & ~CFS_PAGE_MASK);
497 index = pos >> CFS_PAGE_SHIFT;
498 bytes = CFS_PAGE_SIZE - offset;
502 /* prevent read beyond file range */
503 if (/* local_lock && */
504 cmd == OBD_BRW_READ && pos + bytes >= st->st_size) {
505 if (pos >= st->st_size)
507 bytes = st->st_size - pos;
510 /* prepare page for this index */
512 page->addr = buf - offset;
514 page->_offset = offset;
515 page->_count = bytes;
523 group->lig_rwcount += bytes;
527 group->lig_npages += npages;
529 for (i = 0, page = pages; i < npages;
530 i++, page++, llap++, llap_cookie += llap_cookie_size){
531 llap->llap_magic = LLAP_MAGIC;
532 llap->llap_cookie = llap_cookie;
533 rc = obd_prep_async_page(exp, lsm, NULL, page,
534 (obd_off)page->index << CFS_PAGE_SHIFT,
536 llap, &llap->llap_cookie);
539 llap->llap_cookie = NULL;
543 CDEBUG(D_CACHE, "llap %p page %p group %p obj off "LPU64"\n",
544 llap, page, llap->llap_cookie,
545 (obd_off)pages->index << CFS_PAGE_SHIFT);
546 page->private = (unsigned long)llap;
547 llap->llap_page = page;
548 llap->llap_inode = group->lig_inode;
550 rc = obd_queue_group_io(exp, lsm, NULL, group->lig_oig,
551 llap->llap_cookie, cmd,
552 page->_offset, page->_count,
553 group->lig_params->lrp_brw_flags,
554 ASYNC_READY | ASYNC_URGENT |
555 ASYNC_COUNT_STABLE | ASYNC_GROUP_SYNC);
556 if (!local_lock && cmd == OBD_BRW_READ) {
558 * In OST-side locking case short reads cannot be
561 * The root of the problem is that
563 * kms = lov_merge_size(lsm, 1);
565 * glimpse_size(inode);
569 * logic in the read code (both llite and liblustre)
570 * only works correctly when client holds DLM lock on
571 * [start, end]. Without DLM lock KMS can be
572 * completely out of date, and client can either make
573 * spurious short-read (missing concurrent write), or
574 * return stale data (missing concurrent
575 * truncate). For llite client this is fatal, because
576 * incorrect data are cached and can be later sent
577 * back to the server (vide bug 5047). This is hard to
578 * fix by handling short-reads on the server, as there
579 * is no easy way to communicate file size (or amount
580 * of bytes read/written) back to the client,
581 * _especially_ because OSC pages can be sliced and
582 * dices into multiple RPCs arbitrary. Fortunately,
583 * liblustre doesn't cache data and the worst case is
584 * that we get race with concurrent write or truncate.
592 llap->llap_queued = 1;
599 struct llu_io_group * get_io_group(struct inode *inode, int maxpages,
600 struct lustre_rw_params *params)
602 struct llu_io_group *group;
605 if (!llap_cookie_size)
606 llap_cookie_size = obd_prep_async_page(llu_i2obdexp(inode),
610 OBD_ALLOC(group, LLU_IO_GROUP_SIZE(maxpages));
612 return ERR_PTR(-ENOMEM);
615 group->lig_inode = inode;
616 group->lig_maxpages = maxpages;
617 group->lig_params = params;
618 group->lig_llaps = (struct ll_async_page *)(group + 1);
619 group->lig_pages = (struct page *)(&group->lig_llaps[maxpages]);
620 group->lig_llap_cookies = (void *)(&group->lig_pages[maxpages]);
622 rc = oig_init(&group->lig_oig);
624 OBD_FREE(group, LLU_IO_GROUP_SIZE(maxpages));
631 static int max_io_pages(ssize_t len, int iovlen)
633 return (((len + CFS_PAGE_SIZE -1) / CFS_PAGE_SIZE) + 2 + iovlen - 1);
637 void put_io_group(struct llu_io_group *group)
639 struct lov_stripe_md *lsm = llu_i2info(group->lig_inode)->lli_smd;
640 struct obd_export *exp = llu_i2obdexp(group->lig_inode);
641 struct ll_async_page *llap = group->lig_llaps;
644 for (i = 0; i < group->lig_npages; i++, llap++) {
645 if (llap->llap_cookie)
646 obd_teardown_async_page(exp, lsm, NULL,
650 I_RELE(group->lig_inode);
652 oig_release(group->lig_oig);
653 OBD_FREE(group, LLU_IO_GROUP_SIZE(group->lig_maxpages));
657 ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen,
658 _SYSIO_OFF_T pos, ssize_t len,
661 struct llu_io_session *session = (struct llu_io_session *) private;
662 struct inode *inode = session->lis_inode;
663 struct llu_inode_info *lli = llu_i2info(inode);
664 struct intnl_stat *st = llu_i2stat(inode);
665 struct ll_file_data *fd = lli->lli_file_data;
666 struct lustre_handle lockh = {0};
667 struct lov_stripe_md *lsm = lli->lli_smd;
668 struct obd_export *exp = NULL;
669 struct llu_io_group *iogroup;
670 struct lustre_rw_params p;
673 int err, is_read, iovidx, ret;
675 ssize_t ret_len = len;
678 /* in a large iov read/write we'll be repeatedly called.
679 * so give a chance to answer cancel ast here
681 liblustre_wait_event(0);
683 exp = llu_i2obdexp(inode);
687 if (len == 0 || iovlen == 0)
690 if (pos + len > lli->lli_maxbytes)
693 lustre_build_lock_params(session->lis_cmd, lli->lli_open_flags,
694 lli->lli_sbi->ll_lco.lco_flags,
697 iogroup = get_io_group(inode, max_io_pages(len, iovlen), &p);
699 RETURN(PTR_ERR(iogroup));
701 local_lock = p.lrp_lock_mode != LCK_NL;
703 err = llu_extent_lock(fd, inode, lsm, p.lrp_lock_mode, &p.lrp_policy,
704 &lockh, p.lrp_ast_flags);
708 is_read = (session->lis_cmd == OBD_BRW_READ);
711 * If OST-side locking is used, KMS can be completely out of
712 * date, and, hence, cannot be used for short-read
713 * detection. Rely in OST to handle short reads in that case.
715 inode_init_lvb(inode, &lvb);
716 obd_merge_lvb(exp, lsm, &lvb, 1);
718 /* extent.end is last byte of the range */
719 if (p.lrp_policy.l_extent.end >= kms) {
720 /* A glimpse is necessary to determine whether
721 * we return a short read or some zeroes at
722 * the end of the buffer
724 * In the case of OST-side locking KMS can be
725 * completely out of date and short-reads maybe
726 * mishandled. See llu_queue_pio() for more detailed
729 if ((err = llu_glimpse_size(inode))) {
730 GOTO(err_unlock, err);
735 } else if (lli->lli_open_flags & O_APPEND) {
739 for (iovidx = 0; iovidx < iovlen; iovidx++) {
740 char *buf = (char *) iovec[iovidx].iov_base;
741 size_t count = iovec[iovidx].iov_len;
747 if (IS_BAD_PTR(buf) || IS_BAD_PTR(buf + count)) {
748 GOTO(err_unlock, err = -EFAULT);
752 if (/* local_lock && */ pos >= st->st_size)
755 if (pos >= lli->lli_maxbytes) {
756 GOTO(err_unlock, err = -EFBIG);
758 if (pos + count >= lli->lli_maxbytes)
759 count = lli->lli_maxbytes - pos;
762 ret = llu_queue_pio(session->lis_cmd, iogroup, buf, count, pos);
764 GOTO(err_unlock, err = ret);
768 LASSERT(ret == count);
769 obd_adjust_kms(exp, lsm, pos, 0);
770 /* file size grow immediately */
771 if (pos > st->st_size)
779 LASSERT(len == 0 || is_read); /* libsysio should guarantee this */
781 err = obd_trigger_group_io(exp, lsm, NULL, iogroup->lig_oig);
783 GOTO(err_unlock, err);
785 err = oig_wait(iogroup->lig_oig);
787 CERROR("%s error: %s\n", is_read ? "read" : "write", strerror(-err));
788 GOTO(err_unlock, err);
791 ret = llu_extent_unlock(fd, inode, lsm, p.lrp_lock_mode, &lockh);
793 CERROR("extent unlock error %d\n", ret);
795 session->lis_groups[session->lis_ngroups++] = iogroup;
799 llu_extent_unlock(fd, inode, lsm, p.lrp_lock_mode, &lockh);
801 put_io_group(iogroup);
802 RETURN((ssize_t)err);
806 struct llu_io_session *get_io_session(struct inode *ino, int ngroups, int cmd)
808 struct llu_io_session *session;
810 OBD_ALLOC(session, LLU_IO_SESSION_SIZE(ngroups));
815 session->lis_inode = ino;
816 session->lis_max_groups = ngroups;
817 session->lis_cmd = cmd;
821 static void put_io_session(struct llu_io_session *session)
825 for (i = 0; i < session->lis_ngroups; i++) {
826 if (session->lis_groups[i]) {
827 put_io_group(session->lis_groups[i]);
828 session->lis_groups[i] = NULL;
832 I_RELE(session->lis_inode);
833 OBD_FREE(session, LLU_IO_SESSION_SIZE(session->lis_max_groups));
836 static int llu_file_rwx(struct inode *ino,
840 struct llu_io_session *session;
842 int cmd = read ? OBD_BRW_READ : OBD_BRW_WRITE;
845 LASSERT(ioctx->ioctx_xtvlen >= 0);
846 LASSERT(ioctx->ioctx_iovlen >= 0);
848 liblustre_wait_event(0);
850 if (!ioctx->ioctx_xtvlen)
853 /* XXX consider other types later */
854 if (S_ISDIR(llu_i2stat(ino)->st_mode))
856 if (!S_ISREG(llu_i2stat(ino)->st_mode))
859 session = get_io_session(ino, ioctx->ioctx_xtvlen * 2, cmd);
863 cc = _sysio_enumerate_extents(ioctx->ioctx_xtv, ioctx->ioctx_xtvlen,
864 ioctx->ioctx_iov, ioctx->ioctx_iovlen,
865 llu_file_prwv, session);
868 LASSERT(!ioctx->ioctx_cc);
869 ioctx->ioctx_private = session;
872 put_io_session(session);
875 liblustre_wait_event(0);
879 int llu_iop_read(struct inode *ino,
883 struct intnl_stat *st = llu_i2stat(ino);
884 st->st_atime = CURRENT_TIME;
886 return llu_file_rwx(ino, ioctx, 1);
889 int llu_iop_write(struct inode *ino,
892 struct intnl_stat *st = llu_i2stat(ino);
893 st->st_mtime = st->st_ctime = CURRENT_TIME;
895 return llu_file_rwx(ino, ioctx, 0);
898 int llu_iop_iodone(struct ioctx *ioctx)
900 struct llu_io_session *session;
901 struct llu_io_group *group;
902 int i, err = 0, rc = 0;
905 liblustre_wait_event(0);
907 session = (struct llu_io_session *) ioctx->ioctx_private;
909 LASSERT(!IS_ERR(session));
911 for (i = 0; i < session->lis_ngroups; i++) {
912 group = session->lis_groups[i];
915 err = oig_wait(group->lig_oig);
920 ioctx->ioctx_cc += group->lig_rwcount;
922 session->lis_groups[i] = NULL;
928 ioctx->ioctx_cc = -1;
929 ioctx->ioctx_errno = -rc;
932 put_io_session(session);
933 ioctx->ioctx_private = NULL;
934 liblustre_wait_event(0);