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 (c) 2003, 2010, Oracle and/or its affiliates. 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/liblustre/rw.c
38 * Lustre Light block IO
41 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <sys/types.h>
49 #include <sys/queue.h>
53 #include "llite_lib.h"
55 typedef ssize_t llu_file_piov_t(const struct iovec *iovec, int iovlen,
56 _SYSIO_OFF_T pos, ssize_t len,
59 size_t llap_cookie_size;
61 static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock)
63 struct llu_inode_info *lli = llu_i2info(inode);
64 struct lov_stripe_md *lsm = lli->lli_smd;
65 struct obd_export *exp = llu_i2obdexp(inode);
68 struct ldlm_lock *lock;
69 } key = { .name = KEY_LOCK_TO_STRIPE, .lock = lock };
70 __u32 stripe, vallen = sizeof(stripe);
74 if (lsm->lsm_stripe_count == 1)
77 /* get our offset in the lov */
78 rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe, lsm);
80 CERROR("obd_get_info: rc = %d\n", rc);
83 LASSERT(stripe < lsm->lsm_stripe_count);
87 int llu_extent_lock_cancel_cb(struct ldlm_lock *lock,
88 struct ldlm_lock_desc *new, void *data,
91 struct lustre_handle lockh = { 0 };
95 if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
96 LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
101 case LDLM_CB_BLOCKING:
102 ldlm_lock2handle(lock, &lockh);
103 rc = ldlm_cli_cancel(&lockh);
105 CERROR("ldlm_cli_cancel failed: %d\n", rc);
107 case LDLM_CB_CANCELING: {
109 struct llu_inode_info *lli;
110 struct lov_stripe_md *lsm;
114 /* This lock wasn't granted, don't try to evict pages */
115 if (lock->l_req_mode != lock->l_granted_mode)
118 inode = llu_inode_from_lock(lock);
121 lli= llu_i2info(inode);
128 stripe = llu_lock_to_stripe_offset(inode, lock);
129 lock_res_and_lock(lock);
130 kms = ldlm_extent_shift_kms(lock,
131 lsm->lsm_oinfo[stripe]->loi_kms);
132 unlock_res_and_lock(lock);
133 if (lsm->lsm_oinfo[stripe]->loi_kms != kms)
134 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
135 lsm->lsm_oinfo[stripe]->loi_kms, kms);
136 loi_kms_set(lsm->lsm_oinfo[stripe], kms);
148 static int llu_glimpse_callback(struct ldlm_lock *lock, void *reqp)
150 struct ptlrpc_request *req = reqp;
151 struct inode *inode = llu_inode_from_lock(lock);
152 struct llu_inode_info *lli;
158 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
159 lli = llu_i2info(inode);
161 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
162 if (lli->lli_smd == NULL)
163 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
165 /* First, find out which stripe index this lock corresponds to. */
166 if (lli->lli_smd->lsm_stripe_count > 1)
167 stripe = llu_lock_to_stripe_offset(inode, lock);
169 req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
170 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
172 rc = req_capsule_server_pack(&req->rq_pill);
174 CERROR("failed pack reply: %d\n", rc);
178 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
179 lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms;
181 LDLM_DEBUG(lock, "i_size: "LPU64" -> stripe number %u -> kms "LPU64,
182 (__u64)llu_i2stat(inode)->st_size, stripe,lvb->lvb_size);
186 /* These errors are normal races, so we don't want to fill the console
187 * with messages by calling ptlrpc_error() */
188 if (rc == -ELDLM_NO_LOCK_DATA)
189 lustre_pack_reply(req, 1, NULL, NULL);
195 int llu_merge_lvb(struct inode *inode)
197 struct llu_inode_info *lli = llu_i2info(inode);
198 struct llu_sb_info *sbi = llu_i2sbi(inode);
199 struct intnl_stat *st = llu_i2stat(inode);
204 lov_stripe_lock(lli->lli_smd);
205 inode_init_lvb(inode, &lvb);
206 /* merge timestamps the most resently obtained from mds with
207 timestamps obtained from osts */
208 lvb.lvb_atime = lli->lli_lvb.lvb_atime;
209 lvb.lvb_mtime = lli->lli_lvb.lvb_mtime;
210 lvb.lvb_ctime = lli->lli_lvb.lvb_ctime;
211 rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
212 st->st_size = lvb.lvb_size;
213 st->st_blocks = lvb.lvb_blocks;
214 /* handle st_blocks overflow gracefully */
215 if (st->st_blocks < lvb.lvb_blocks)
216 st->st_blocks = ~0UL;
217 st->st_mtime = lvb.lvb_mtime;
218 st->st_atime = lvb.lvb_atime;
219 st->st_ctime = lvb.lvb_ctime;
220 lov_stripe_unlock(lli->lli_smd);
225 int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
226 struct lov_stripe_md *lsm, int mode,
227 ldlm_policy_data_t *policy, struct lustre_handle *lockh,
230 struct llu_sb_info *sbi = llu_i2sbi(inode);
231 struct intnl_stat *st = llu_i2stat(inode);
232 struct ldlm_enqueue_info einfo = { 0 };
233 struct obd_info oinfo = { { { 0 } } };
238 LASSERT(!lustre_handle_is_used(lockh));
239 CLASSERT(ELDLM_OK == 0);
241 /* XXX phil: can we do this? won't it screw the file size up? */
242 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
243 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
246 CDEBUG(D_DLMTRACE, "Locking inode "LPU64", start "LPU64" end "LPU64"\n",
247 (__u64)st->st_ino, policy->l_extent.start,
248 policy->l_extent.end);
250 einfo.ei_type = LDLM_EXTENT;
251 einfo.ei_mode = mode;
252 einfo.ei_cb_bl = llu_extent_lock_cancel_cb;
253 einfo.ei_cb_cp = ldlm_completion_ast;
254 einfo.ei_cb_gl = llu_glimpse_callback;
255 einfo.ei_cbdata = inode;
257 oinfo.oi_policy = *policy;
258 oinfo.oi_lockh = lockh;
260 oinfo.oi_flags = ast_flags;
262 rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL);
263 *policy = oinfo.oi_policy;
267 inode_init_lvb(inode, &lvb);
268 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1);
269 if (policy->l_extent.start == 0 &&
270 policy->l_extent.end == OBD_OBJECT_EOF)
271 st->st_size = lvb.lvb_size;
274 st->st_mtime = lvb.lvb_mtime;
275 st->st_atime = lvb.lvb_atime;
276 st->st_ctime = lvb.lvb_ctime;
282 int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode,
283 struct lov_stripe_md *lsm, int mode,
284 struct lustre_handle *lockh)
286 struct llu_sb_info *sbi = llu_i2sbi(inode);
290 CLASSERT(ELDLM_OK == 0);
292 /* XXX phil: can we do this? won't it screw the file size up? */
293 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
294 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
297 rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh);
303 ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen,
304 _SYSIO_OFF_T pos, ssize_t len,
307 struct llu_io_session *session = (struct llu_io_session *) private;
308 struct inode *inode = session->lis_inode;
309 struct llu_inode_info *lli = llu_i2info(inode);
317 /* in a large iov read/write we'll be repeatedly called.
318 * so give a chance to answer cancel ast here
320 liblustre_wait_event(0);
322 if (len == 0 || iovlen == 0)
325 if (pos + len > lli->lli_maxbytes)
328 env = cl_env_get(&refcheck);
330 RETURN(PTR_ERR(env));
332 io = ccc_env_thread_io(env);
333 if (cl_io_rw_init(env, io, session->lis_cmd == OBD_BRW_WRITE?CIT_WRITE:
337 sio = slp_env_io(env);
338 cio = ccc_env_io(env);
339 /* XXX this is not right: cio->cui_iov can be modified. */
340 cio->cui_iov = (struct iovec *)iovec;
341 cio->cui_nrsegs = iovlen;
342 cio->cui_tot_nrsegs = iovlen;
343 sio->sio_session = session;
344 err = cl_io_loop(env, io);
350 cl_env_put(env, &refcheck);
359 struct llu_io_session *get_io_session(struct inode *ino, int ngroups, int cmd)
361 struct llu_io_session *session;
363 OBD_ALLOC_PTR(session);
368 session->lis_inode = ino;
369 session->lis_max_groups = ngroups;
370 session->lis_cmd = cmd;
374 static void put_io_session(struct llu_io_session *session)
376 I_RELE(session->lis_inode);
377 OBD_FREE_PTR(session);
380 static int llu_file_rwx(struct inode *ino,
384 struct llu_io_session *session;
386 int cmd = read ? OBD_BRW_READ : OBD_BRW_WRITE;
389 LASSERT(ioctx->ioctx_xtvlen >= 0);
390 LASSERT(ioctx->ioctx_iovlen >= 0);
392 liblustre_wait_event(0);
394 if (!ioctx->ioctx_xtvlen)
397 /* XXX consider other types later */
398 if (S_ISDIR(llu_i2stat(ino)->st_mode))
400 if (!S_ISREG(llu_i2stat(ino)->st_mode))
403 session = get_io_session(ino, ioctx->ioctx_xtvlen * 2, cmd);
407 cc = _sysio_enumerate_extents(ioctx->ioctx_xtv, ioctx->ioctx_xtvlen,
408 ioctx->ioctx_iov, ioctx->ioctx_iovlen,
409 llu_file_prwv, session);
412 LASSERT(!ioctx->ioctx_cc);
413 ioctx->ioctx_private = session;
416 put_io_session(session);
419 liblustre_wait_event(0);
423 void llu_io_init(struct cl_io *io, struct inode *inode, int write)
425 struct llu_inode_info *lli = llu_i2info(inode);
427 io->u.ci_rw.crw_nonblock = lli->lli_open_flags & O_NONBLOCK;
429 io->u.ci_wr.wr_append = lli->lli_open_flags & O_APPEND;
430 io->ci_obj = llu_i2info(inode)->lli_clob;
432 if ((lli->lli_open_flags & O_APPEND) && write)
433 io->ci_lockreq = CILR_MANDATORY;
435 io->ci_lockreq = CILR_NEVER;
438 int llu_iop_read(struct inode *ino,
441 struct intnl_stat *st = llu_i2stat(ino);
448 st->st_atime = CFS_CURRENT_TIME;
450 env = cl_env_get(&refcheck);
452 RETURN(PTR_ERR(env));
454 io = ccc_env_thread_io(env);
455 llu_io_init(io, ino, 0);
457 ret = llu_file_rwx(ino, ioctx, 1);
459 cl_env_put(env, &refcheck);
463 int llu_iop_write(struct inode *ino,
466 struct intnl_stat *st = llu_i2stat(ino);
472 st->st_mtime = st->st_ctime = CFS_CURRENT_TIME;
474 env = cl_env_get(&refcheck);
476 RETURN(PTR_ERR(env));
478 io = ccc_env_thread_io(env);
479 llu_io_init(io, ino, 1);
481 ret = llu_file_rwx(ino, ioctx, 0);
482 cl_env_put(env, &refcheck);
486 int llu_iop_iodone(struct ioctx *ioctx)
488 struct llu_io_session *session;
494 liblustre_wait_event(0);
496 env = cl_env_get(&refcheck);
498 RETURN(PTR_ERR(env));
500 io = &ccc_env_info(env)->cti_io;
502 cl_env_put(env, &refcheck);
503 session = (struct llu_io_session *) ioctx->ioctx_private;
505 LASSERT(!IS_ERR(session));
507 if (session->lis_rc == 0) {
508 ioctx->ioctx_cc = session->lis_rwcount;
510 LASSERT(session->lis_rc < 0);
511 ioctx->ioctx_cc = -1;
512 ioctx->ioctx_errno = -session->lis_rc;
515 put_io_session(session);
516 ioctx->ioctx_private = NULL;
517 liblustre_wait_event(0);