X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fliblustre%2Frw.c;h=7fdf5aefb54bd90cbebaeef04e1d11f5a71d48fc;hp=c6f4bd0f5625bbb890f5983d46a0168c7cce8b82;hb=422e5995808bee60407a78020825ab369e1449bd;hpb=fbb7ead129258897f5a5d5c9ce28d31fbbe5bca2 diff --git a/lustre/liblustre/rw.c b/lustre/liblustre/rw.c index c6f4bd0..7fdf5ae 100644 --- a/lustre/liblustre/rw.c +++ b/lustre/liblustre/rw.c @@ -1,24 +1,41 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Lustre Light block IO + * GPL HEADER START + * + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * Copyright (c) 2002-2004 Cluster File Systems, Inc. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * - * This file is part of Lustre, http://www.lustre.org. + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. * - * Lustre is free software; you can redistribute it and/or - * modify it under the terms of version 2 of the GNU General Public - * License as published by the Free Software Foundation. + * GPL HEADER END + */ +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Use is subject to license terms. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. * - * Lustre is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * lustre/liblustre/rw.c * - * You should have received a copy of the GNU General Public License - * along with Lustre; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * Lustre Light block IO */ #define DEBUG_SUBSYSTEM S_LLITE @@ -28,14 +45,16 @@ #include #include #include +#include #include #include #include +#include #ifdef HAVE_XTIO_H #include #endif -#include +#include #include #include #ifdef HAVE_FILE_H @@ -46,48 +65,21 @@ #include "llite_lib.h" -struct llu_io_group -{ - struct obd_io_group *lig_oig; - struct inode *lig_inode; - int lig_maxpages; - int lig_npages; - __u64 lig_rwcount; - struct ll_async_page *lig_llap; - struct page *lig_pages; -}; - -#define LLU_IO_GROUP_SIZE(x) \ - (sizeof(struct llu_io_group) + \ - sizeof(struct ll_async_page) * (x) + \ - sizeof(struct page) * (x)) - -struct llu_io_session -{ - struct inode *lis_inode; - int lis_cmd; - int lis_max_groups; - int lis_ngroups; - struct llu_io_group *lis_groups[0]; -}; -#define LLU_IO_SESSION_SIZE(x) \ - (sizeof(struct llu_io_session) + (x) * 2 * sizeof(void *)) - - typedef ssize_t llu_file_piov_t(const struct iovec *iovec, int iovlen, _SYSIO_OFF_T pos, ssize_t len, void *private); +size_t llap_cookie_size; + static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock) { struct llu_inode_info *lli = llu_i2info(inode); struct lov_stripe_md *lsm = lli->lli_smd; - struct obd_export *exp = llu_i2dtexp(inode); + struct obd_export *exp = llu_i2obdexp(inode); struct { char name[16]; struct ldlm_lock *lock; - struct lov_stripe_md *lsm; - } key = { .name = "lock_to_stripe", .lock = lock, .lsm = lsm }; + } key = { .name = KEY_LOCK_TO_STRIPE, .lock = lock }; __u32 stripe, vallen = sizeof(stripe); int rc; ENTRY; @@ -96,7 +88,7 @@ static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock RETURN(0); /* get our offset in the lov */ - rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe); + rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe, lsm); if (rc != 0) { CERROR("obd_get_info: rc = %d\n", rc); LBUG(); @@ -105,9 +97,9 @@ static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock RETURN(stripe); } -static int llu_extent_lock_callback(struct ldlm_lock *lock, - struct ldlm_lock_desc *new, void *data, - int flag) +int llu_extent_lock_cancel_cb(struct ldlm_lock *lock, + struct ldlm_lock_desc *new, void *data, + int flag) { struct lustre_handle lockh = { 0 }; int rc; @@ -147,14 +139,14 @@ static int llu_extent_lock_callback(struct ldlm_lock *lock, lsm = lli->lli_smd; stripe = llu_lock_to_stripe_offset(inode, lock); - l_lock(&lock->l_resource->lr_namespace->ns_lock); + lock_res_and_lock(lock); kms = ldlm_extent_shift_kms(lock, - lsm->lsm_oinfo[stripe].loi_kms); - l_unlock(&lock->l_resource->lr_namespace->ns_lock); - if (lsm->lsm_oinfo[stripe].loi_kms != kms) + lsm->lsm_oinfo[stripe]->loi_kms); + unlock_res_and_lock(lock); + if (lsm->lsm_oinfo[stripe]->loi_kms != kms) LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64, - lsm->lsm_oinfo[stripe].loi_kms, kms); - lsm->lsm_oinfo[stripe].loi_kms = kms; + lsm->lsm_oinfo[stripe]->loi_kms, kms); + loi_kms_set(lsm->lsm_oinfo[stripe], kms); iput: I_RELE(inode); break; @@ -172,7 +164,7 @@ static int llu_glimpse_callback(struct ldlm_lock *lock, void *reqp) struct inode *inode = llu_inode_from_lock(lock); struct llu_inode_info *lli; struct ost_lvb *lvb; - int rc, size = sizeof(*lvb), stripe = 0; + int rc, stripe = 0; ENTRY; if (inode == NULL) @@ -187,63 +179,51 @@ static int llu_glimpse_callback(struct ldlm_lock *lock, void *reqp) if (lli->lli_smd->lsm_stripe_count > 1) stripe = llu_lock_to_stripe_offset(inode, lock); - rc = lustre_pack_reply(req, 1, &size, NULL); + req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK); + req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, + sizeof(*lvb)); + rc = req_capsule_server_pack(&req->rq_pill); if (rc) { - CERROR("lustre_pack_reply: %d\n", rc); + CERROR("failed pack reply: %d\n", rc); GOTO(iput, rc); } - lvb = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*lvb)); - lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe].loi_kms; + lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB); + lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms; LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64, - lli->lli_st_size, stripe, lvb->lvb_size); + (__u64)llu_i2stat(inode)->st_size, stripe,lvb->lvb_size); iput: I_RELE(inode); out: /* These errors are normal races, so we don't want to fill the console * with messages by calling ptlrpc_error() */ if (rc == -ELDLM_NO_LOCK_DATA) - lustre_pack_reply(req, 0, NULL, NULL); + lustre_pack_reply(req, 1, NULL, NULL); req->rq_status = rc; return rc; } -__u64 lov_merge_size(struct lov_stripe_md *lsm, int kms_only); -__u64 lov_merge_blocks(struct lov_stripe_md *lsm); -__u64 lov_merge_mtime(struct lov_stripe_md *lsm, __u64 current_time); - -/* NB: lov_merge_size will prefer locally cached writes if they extend the - * file (because it prefers KMS over RSS when larger) */ -int llu_glimpse_size(struct inode *inode) +int llu_merge_lvb(struct inode *inode) { struct llu_inode_info *lli = llu_i2info(inode); struct llu_sb_info *sbi = llu_i2sbi(inode); - ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } }; - struct lustre_handle lockh = { 0 }; - int rc, flags = LDLM_FL_HAS_INTENT; + struct intnl_stat *st = llu_i2stat(inode); + struct ost_lvb lvb; + int rc; ENTRY; - CDEBUG(D_DLMTRACE, "Glimpsing inode %lu\n", lli->lli_st_ino); - - rc = obd_enqueue(sbi->ll_dt_exp, lli->lli_smd, LDLM_EXTENT, &policy, - LCK_PR, &flags, llu_extent_lock_callback, - ldlm_completion_ast, llu_glimpse_callback, inode, - sizeof(struct ost_lvb), lustre_swab_ost_lvb, &lockh); - if (rc) { - CERROR("obd_enqueue returned rc %d, returning -EIO\n", rc); - RETURN(rc > 0 ? -EIO : rc); - } - - lli->lli_st_size = lov_merge_size(lli->lli_smd, 0); - lli->lli_st_blocks = lov_merge_blocks(lli->lli_smd); - //lli->lli_st_mtime = lov_merge_mtime(lli->lli_smd, inode->i_mtime); - - CDEBUG(D_DLMTRACE, "glimpse: size: %llu, blocks: %lu\n", - lli->lli_st_size, lli->lli_st_blocks); - - obd_cancel(sbi->ll_dt_exp, lli->lli_smd, LCK_PR, &lockh); + inode_init_lvb(inode, &lvb); + rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0); + st->st_size = lvb.lvb_size; + st->st_blocks = lvb.lvb_blocks; + /* handle st_blocks overflow gracefully */ + if (st->st_blocks < lvb.lvb_blocks) + st->st_blocks = ~0UL; + st->st_mtime = lvb.lvb_mtime; + st->st_atime = lvb.lvb_atime; + st->st_ctime = lvb.lvb_ctime; RETURN(rc); } @@ -254,32 +234,53 @@ int llu_extent_lock(struct ll_file_data *fd, struct inode *inode, int ast_flags) { struct llu_sb_info *sbi = llu_i2sbi(inode); - struct llu_inode_info *lli = llu_i2info(inode); + struct intnl_stat *st = llu_i2stat(inode); + struct ldlm_enqueue_info einfo = { 0 }; + struct obd_info oinfo = { { { 0 } } }; + struct ost_lvb lvb; int rc; ENTRY; - LASSERT(lockh->cookie == 0); + LASSERT(!lustre_handle_is_used(lockh)); + CLASSERT(ELDLM_OK == 0); /* XXX phil: can we do this? won't it screw the file size up? */ if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) || - (sbi->ll_flags & LL_SBI_NOLCK)) + (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL) RETURN(0); - CDEBUG(D_DLMTRACE, "Locking inode %lu, start "LPU64" end "LPU64"\n", - lli->lli_st_ino, policy->l_extent.start, policy->l_extent.end); + CDEBUG(D_DLMTRACE, "Locking inode %llu, start "LPU64" end "LPU64"\n", + (__u64)st->st_ino, policy->l_extent.start, + policy->l_extent.end); + + einfo.ei_type = LDLM_EXTENT; + einfo.ei_mode = mode; + einfo.ei_cb_bl = llu_extent_lock_cancel_cb; + einfo.ei_cb_cp = ldlm_completion_ast; + einfo.ei_cb_gl = llu_glimpse_callback; + einfo.ei_cbdata = inode; - rc = obd_enqueue(sbi->ll_dt_exp, lsm, LDLM_EXTENT, policy, mode, - &ast_flags, llu_extent_lock_callback, - ldlm_completion_ast, llu_glimpse_callback, inode, - sizeof(struct ost_lvb), lustre_swab_ost_lvb, lockh); + oinfo.oi_policy = *policy; + oinfo.oi_lockh = lockh; + oinfo.oi_md = lsm; + oinfo.oi_flags = ast_flags; + + rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL); + *policy = oinfo.oi_policy; if (rc > 0) rc = -EIO; + inode_init_lvb(inode, &lvb); + obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1); if (policy->l_extent.start == 0 && policy->l_extent.end == OBD_OBJECT_EOF) - lli->lli_st_size = lov_merge_size(lsm, 1); + st->st_size = lvb.lvb_size; - //inode->i_mtime = lov_merge_mtime(lsm, inode->i_mtime); + if (rc == 0) { + st->st_mtime = lvb.lvb_mtime; + st->st_atime = lvb.lvb_atime; + st->st_ctime = lvb.lvb_ctime; + } RETURN(rc); } @@ -292,9 +293,11 @@ int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode, int rc; ENTRY; + CLASSERT(ELDLM_OK == 0); + /* XXX phil: can we do this? won't it screw the file size up? */ if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) || - (sbi->ll_flags & LL_SBI_NOLCK)) + (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL) RETURN(0); rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh); @@ -302,197 +305,6 @@ int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode, RETURN(rc); } -#define LLAP_MAGIC 12346789 - -struct ll_async_page { - int llap_magic; - void *llap_cookie; - int llap_queued; - struct page *llap_page; - struct inode *llap_inode; -}; - -static void llu_ap_fill_obdo(void *data, int cmd, struct obdo *oa) -{ - struct ll_async_page *llap; - struct inode *inode; - struct lov_stripe_md *lsm; - obd_valid valid_flags; - ENTRY; - - llap = LLAP_FROM_COOKIE(data); - inode = llap->llap_inode; - lsm = llu_i2info(inode)->lli_smd; - - oa->o_id = lsm->lsm_object_id; - oa->o_valid = OBD_MD_FLID; - valid_flags = OBD_MD_FLTYPE | OBD_MD_FLATIME; - if (cmd == OBD_BRW_WRITE) - valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME; - - obdo_from_inode(oa, inode, valid_flags); - EXIT; -} - -/* called for each page in a completed rpc.*/ -static void llu_ap_completion(void *data, int cmd, struct obdo *oa, int rc) -{ - struct ll_async_page *llap; - struct page *page; - - llap = LLAP_FROM_COOKIE(data); - llap->llap_queued = 0; - page = llap->llap_page; - - if (rc != 0) { - if (cmd == OBD_BRW_WRITE) - CERROR("writeback error on page %p index %ld: %d\n", - page, page->index, rc); - } - EXIT; -} - -static struct obd_async_page_ops llu_async_page_ops = { - .ap_make_ready = NULL, - .ap_refresh_count = NULL, - .ap_fill_obdo = llu_ap_fill_obdo, - .ap_completion = llu_ap_completion, -}; - -static int llu_queue_pio(int cmd, struct llu_io_group *group, - char *buf, size_t count, loff_t pos) -{ - struct llu_inode_info *lli = llu_i2info(group->lig_inode); - struct lov_stripe_md *lsm = lli->lli_smd; - struct obd_export *exp = llu_i2dtexp(group->lig_inode); - struct page *pages = &group->lig_pages[group->lig_npages]; - struct ll_async_page *llap = &group->lig_llap[group->lig_npages]; - int i, rc, npages = 0, ret_bytes = 0; - ENTRY; - - if (!exp) - RETURN(-EINVAL); - - /* prepare the pages array */ - do { - unsigned long index, offset, bytes; - - offset = (pos & ~PAGE_CACHE_MASK); - index = pos >> PAGE_CACHE_SHIFT; - bytes = PAGE_CACHE_SIZE - offset; - if (bytes > count) - bytes = count; - - /* prevent read beyond file range */ - if ((cmd == OBD_BRW_READ) && - (pos + bytes) >= lli->lli_st_size) { - if (pos >= lli->lli_st_size) - break; - bytes = lli->lli_st_size - pos; - } - - /* prepare page for this index */ - pages[npages].index = index; - pages[npages].addr = buf - offset; - - pages[npages]._offset = offset; - pages[npages]._count = bytes; - - npages++; - count -= bytes; - pos += bytes; - buf += bytes; - - group->lig_rwcount += bytes; - ret_bytes += bytes; - } while (count); - - group->lig_npages += npages; - - for (i = 0; i < npages; i++) { - llap[i].llap_magic = LLAP_MAGIC; - rc = obd_prep_async_page(exp, lsm, NULL, &pages[i], - (obd_off)pages[i].index << PAGE_SHIFT, - &llu_async_page_ops, - &llap[i], &llap[i].llap_cookie); - if (rc) { - LASSERT(rc < 0); - llap[i].llap_cookie = NULL; - RETURN(rc); - } - CDEBUG(D_CACHE, "llap %p page %p group %p obj off "LPU64"\n", - &llap[i], &pages[i], llap[i].llap_cookie, - (obd_off)pages[i].index << PAGE_SHIFT); - pages[i].private = (unsigned long)&llap[i]; - llap[i].llap_page = &pages[i]; - llap[i].llap_inode = group->lig_inode; - - rc = obd_queue_group_io(exp, lsm, NULL, group->lig_oig, - llap[i].llap_cookie, cmd, - pages[i]._offset, pages[i]._count, 0, - ASYNC_READY | ASYNC_URGENT | - ASYNC_COUNT_STABLE | ASYNC_GROUP_SYNC); - if (rc) { - LASSERT(rc < 0); - RETURN(rc); - } - - llap[i].llap_queued = 1; - } - - RETURN(ret_bytes); -} - -static -struct llu_io_group * get_io_group(struct inode *inode, int maxpages) -{ - struct llu_io_group *group; - int rc; - - OBD_ALLOC(group, LLU_IO_GROUP_SIZE(maxpages)); - if (!group) - return ERR_PTR(-ENOMEM); - - I_REF(inode); - group->lig_inode = inode; - group->lig_maxpages = maxpages; - group->lig_llap = (struct ll_async_page *)(group + 1); - group->lig_pages = (struct page *) (group->lig_llap + maxpages); - - rc = oig_init(&group->lig_oig); - if (rc) { - OBD_FREE(group, LLU_IO_GROUP_SIZE(maxpages)); - return ERR_PTR(rc); - } - - return group; -} - -static int max_io_pages(ssize_t len, int iovlen) -{ - return (((len + PAGE_SIZE -1) / PAGE_SIZE) + 2 + iovlen - 1); -} - -static -void put_io_group(struct llu_io_group *group) -{ - struct lov_stripe_md *lsm = llu_i2info(group->lig_inode)->lli_smd; - struct obd_export *exp = llu_i2dtexp(group->lig_inode); - struct ll_async_page *llap = group->lig_llap; - int i; - - for (i = 0; i< group->lig_npages; i++) { - if (llap[i].llap_cookie) - obd_teardown_async_page(exp, lsm, NULL, - llap[i].llap_cookie); - } - - I_RELE(group->lig_inode); - - oig_release(group->lig_oig); - OBD_FREE(group, LLU_IO_GROUP_SIZE(group->lig_maxpages)); -} - static ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen, _SYSIO_OFF_T pos, ssize_t len, @@ -501,16 +313,11 @@ ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen, struct llu_io_session *session = (struct llu_io_session *) private; struct inode *inode = session->lis_inode; struct llu_inode_info *lli = llu_i2info(inode); - struct ll_file_data *fd = lli->lli_file_data; - struct lustre_handle lockh = {0}; - struct lov_stripe_md *lsm = lli->lli_smd; - struct obd_export *exp = NULL; - ldlm_policy_data_t policy; - struct llu_io_group *iogroup; - int astflag = (lli->lli_open_flags & O_NONBLOCK) ? - LDLM_FL_BLOCK_NOWAIT : 0; - __u64 kms; - int err, is_read, lock_mode, iovidx, ret; + int err; + struct lu_env *env; + struct cl_io *io; + struct slp_io *sio; + int refcheck; ENTRY; /* in a large iov read/write we'll be repeatedly called. @@ -518,114 +325,40 @@ ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen, */ liblustre_wait_event(0); - exp = llu_i2dtexp(inode); - if (exp == NULL) - RETURN(-EINVAL); - if (len == 0 || iovlen == 0) RETURN(0); if (pos + len > lli->lli_maxbytes) RETURN(-ERANGE); - iogroup = get_io_group(inode, max_io_pages(len, iovlen)); - if (IS_ERR(iogroup)) - RETURN(PTR_ERR(iogroup)); - - is_read = session->lis_cmd == OBD_BRW_READ; - lock_mode = is_read ? LCK_PR : LCK_PW; - - if (!is_read && (lli->lli_open_flags & O_APPEND)) { - policy.l_extent.start = 0; - policy.l_extent.end = OBD_OBJECT_EOF; - } else { - policy.l_extent.start = pos; - policy.l_extent.end = pos + len - 1; - } - - err = llu_extent_lock(fd, inode, lsm, lock_mode, &policy, - &lockh, astflag); - if (err != ELDLM_OK) - GOTO(err_put, err); - - if (is_read) { - kms = lov_merge_size(lsm, 1); - if (policy.l_extent.end > kms) { - /* A glimpse is necessary to determine whether we - * return a short read or some zeroes at the end of - * the buffer */ - if ((err = llu_glimpse_size(inode))) { - llu_extent_unlock(fd, inode, lsm, - lock_mode, &lockh); - GOTO(err_put, err); - } - } else { - lli->lli_st_size = kms; - } + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + io = &ccc_env_info(env)->cti_io; + + if (cl_io_rw_init(env, io, session->lis_cmd == OBD_BRW_WRITE?CIT_WRITE: + CIT_READ, + pos, len) == 0) { + struct ccc_io *cio; + sio = slp_env_io(env); + cio = ccc_env_io(env); + /* XXX this is not right: cio->cui_iov can be modified. */ + cio->cui_iov = (struct iovec *)iovec; + cio->cui_nrsegs = iovlen; + sio->sio_session = session; + err = cl_io_loop(env, io); } else { - if (lli->lli_open_flags & O_APPEND) - pos = lli->lli_st_size; - } - - for (iovidx = 0; iovidx < iovlen; iovidx++) { - char *buf = (char *) iovec[iovidx].iov_base; - size_t count = iovec[iovidx].iov_len; - - if (!count) - continue; - if (len < count) - count = len; - if (IS_BAD_PTR(buf) || IS_BAD_PTR(buf + count)) { - llu_extent_unlock(fd, inode, lsm, lock_mode, &lockh); - GOTO(err_put, err = -EFAULT); - } - - if (is_read) { - if (pos >= lli->lli_st_size) - break; - } else { - if (pos >= lli->lli_maxbytes) { - llu_extent_unlock(fd, inode, lsm, lock_mode, - &lockh); - GOTO(err_put, err = -EFBIG); - } - if (pos + count >= lli->lli_maxbytes) - count = lli->lli_maxbytes - pos; - } - - ret = llu_queue_pio(session->lis_cmd, iogroup, buf, count, pos); - if (ret < 0) { - llu_extent_unlock(fd, inode, lsm, lock_mode, &lockh); - GOTO(err_put, err = ret); - } else { - pos += ret; - if (!is_read) { - LASSERT(ret == count); - obd_adjust_kms(exp, lsm, pos, 0); - /* file size grow immediately */ - if (pos > lli->lli_st_size) - lli->lli_st_size = pos; - } - len -= ret; - if (!len) - break; - } + /* XXX WTF? */ + LBUG(); } - LASSERT(len == 0 || is_read); /* libsysio should guarantee this */ - - err = llu_extent_unlock(fd, inode, lsm, lock_mode, &lockh); - if (err) - CERROR("extent unlock error %d\n", err); + cl_io_fini(env, io); + cl_env_put(env, &refcheck); - err = obd_trigger_group_io(exp, lsm, NULL, iogroup->lig_oig); - if (err) - GOTO(err_put, err); + if (err < 0) + RETURN(err); - session->lis_groups[session->lis_ngroups++] = iogroup; - RETURN(0); -err_put: - put_io_group(iogroup); - RETURN((ssize_t)err); + RETURN(len); } static @@ -633,7 +366,7 @@ struct llu_io_session *get_io_session(struct inode *ino, int ngroups, int cmd) { struct llu_io_session *session; - OBD_ALLOC(session, LLU_IO_SESSION_SIZE(ngroups)); + OBD_ALLOC_PTR(session); if (!session) return NULL; @@ -646,17 +379,8 @@ struct llu_io_session *get_io_session(struct inode *ino, int ngroups, int cmd) static void put_io_session(struct llu_io_session *session) { - int i; - - for (i = 0; i < session->lis_ngroups; i++) { - if (session->lis_groups[i]) { - put_io_group(session->lis_groups[i]); - session->lis_groups[i] = NULL; - } - } - I_RELE(session->lis_inode); - OBD_FREE(session, LLU_IO_SESSION_SIZE(session->lis_max_groups)); + OBD_FREE_PTR(session); } static int llu_file_rwx(struct inode *ino, @@ -677,7 +401,10 @@ static int llu_file_rwx(struct inode *ino, RETURN(0); /* XXX consider other types later */ - LASSERT(S_ISREG(llu_i2info(ino)->lli_st_mode)); + if (S_ISDIR(llu_i2stat(ino)->st_mode)) + RETURN(-EISDIR); + if (!S_ISREG(llu_i2stat(ino)->st_mode)) + RETURN(-EOPNOTSUPP); session = get_io_session(ino, ioctx->ioctx_xtvlen * 2, cmd); if (!session) @@ -690,75 +417,112 @@ static int llu_file_rwx(struct inode *ino, if (cc >= 0) { LASSERT(!ioctx->ioctx_cc); ioctx->ioctx_private = session; - RETURN(0); + cc = 0; } else { put_io_session(session); - RETURN(cc); } + + liblustre_wait_event(0); + RETURN(cc); +} + +void llu_io_init(struct cl_io *io, struct inode *inode, int write) +{ + struct llu_inode_info *lli = llu_i2info(inode); + + memset(io, 0, sizeof *io); + + io->u.ci_rw.crw_nonblock = lli->lli_open_flags & O_NONBLOCK; + if (write) + io->u.ci_wr.wr_append = lli->lli_open_flags & O_APPEND; + io->ci_obj = llu_i2info(inode)->lli_clob; + + if ((lli->lli_open_flags & O_APPEND) && write) + io->ci_lockreq = CILR_MANDATORY; + else + io->ci_lockreq = CILR_NEVER; } int llu_iop_read(struct inode *ino, struct ioctx *ioctx) { - return llu_file_rwx(ino, ioctx, 1); + struct intnl_stat *st = llu_i2stat(ino); + struct lu_env *env; + struct cl_io *io; + int refcheck; + int ret; + + /* BUG: 5972 */ + st->st_atime = CURRENT_TIME; + + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + io = &ccc_env_info(env)->cti_io; + llu_io_init(io, ino, 0); + + ret = llu_file_rwx(ino, ioctx, 1); + + cl_env_put(env, &refcheck); + return ret; } int llu_iop_write(struct inode *ino, struct ioctx *ioctx) { - struct iattr iattr; - int rc; + struct intnl_stat *st = llu_i2stat(ino); + struct lu_env *env; + struct cl_io *io; + int refcheck; + int ret; - memset(&iattr, 0, sizeof(iattr)); - iattr.ia_mtime = iattr.ia_atime = CURRENT_TIME; - iattr.ia_valid = ATTR_MTIME | ATTR_ATIME | ATTR_RAW; + st->st_mtime = st->st_ctime = CURRENT_TIME; - liblustre_wait_event(0); - rc = llu_setattr_raw(ino, &iattr); - if (rc) { - CERROR("failed to set mtime/atime during write: %d", rc); - /* XXX should continue or return error? */ - } + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - return llu_file_rwx(ino, ioctx, 0); + io = &ccc_env_info(env)->cti_io; + llu_io_init(io, ino, 1); + + ret = llu_file_rwx(ino, ioctx, 0); + cl_env_put(env, &refcheck); + return ret; } int llu_iop_iodone(struct ioctx *ioctx) { struct llu_io_session *session; - struct llu_io_group *group; - int i, err = 0, rc = 0; + struct lu_env *env; + struct cl_io *io; + int refcheck; ENTRY; liblustre_wait_event(0); + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + io = &ccc_env_info(env)->cti_io; + cl_io_fini(env, io); + cl_env_put(env, &refcheck); session = (struct llu_io_session *) ioctx->ioctx_private; LASSERT(session); LASSERT(!IS_ERR(session)); - for (i = 0; i < session->lis_ngroups; i++) { - group = session->lis_groups[i]; - if (group) { - if (!rc) { - err = oig_wait(group->lig_oig); - if (err) - rc = err; - } - if (!rc) - ioctx->ioctx_cc += group->lig_rwcount; - put_io_group(group); - session->lis_groups[i] = NULL; - } - } - - if (rc) { - LASSERT(rc < 0); + if (session->lis_rc == 0) { + ioctx->ioctx_cc = session->lis_rwcount; + } else { + LASSERT(session->lis_rc < 0); ioctx->ioctx_cc = -1; - ioctx->ioctx_errno = -rc; + ioctx->ioctx_errno = -session->lis_rc; } put_io_session(session); ioctx->ioctx_private = NULL; + liblustre_wait_event(0); RETURN(1); }