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 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/liblustre/rw.c
41 * Lustre Light block IO
44 #define DEBUG_SUBSYSTEM S_LLITE
50 #include <sys/types.h>
52 #include <sys/queue.h>
56 #include "llite_lib.h"
58 typedef ssize_t llu_file_piov_t(const struct iovec *iovec, int iovlen,
59 _SYSIO_OFF_T pos, ssize_t len,
62 size_t llap_cookie_size;
64 static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock)
66 struct llu_inode_info *lli = llu_i2info(inode);
67 struct lov_stripe_md *lsm = lli->lli_smd;
68 struct obd_export *exp = llu_i2obdexp(inode);
71 struct ldlm_lock *lock;
72 } key = { .name = KEY_LOCK_TO_STRIPE, .lock = lock };
73 __u32 stripe, vallen = sizeof(stripe);
77 if (lsm->lsm_stripe_count == 1)
80 /* get our offset in the lov */
81 rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe, lsm);
83 CERROR("obd_get_info: rc = %d\n", rc);
86 LASSERT(stripe < lsm->lsm_stripe_count);
90 int llu_extent_lock_cancel_cb(struct ldlm_lock *lock,
91 struct ldlm_lock_desc *new, void *data,
94 struct lustre_handle lockh = { 0 };
98 if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
99 LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
104 case LDLM_CB_BLOCKING:
105 ldlm_lock2handle(lock, &lockh);
106 rc = ldlm_cli_cancel(&lockh);
108 CERROR("ldlm_cli_cancel failed: %d\n", rc);
110 case LDLM_CB_CANCELING: {
112 struct llu_inode_info *lli;
113 struct lov_stripe_md *lsm;
117 /* This lock wasn't granted, don't try to evict pages */
118 if (lock->l_req_mode != lock->l_granted_mode)
121 inode = llu_inode_from_lock(lock);
124 lli= llu_i2info(inode);
131 stripe = llu_lock_to_stripe_offset(inode, lock);
132 lock_res_and_lock(lock);
133 kms = ldlm_extent_shift_kms(lock,
134 lsm->lsm_oinfo[stripe]->loi_kms);
135 unlock_res_and_lock(lock);
136 if (lsm->lsm_oinfo[stripe]->loi_kms != kms)
137 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
138 lsm->lsm_oinfo[stripe]->loi_kms, kms);
139 loi_kms_set(lsm->lsm_oinfo[stripe], kms);
151 static int llu_glimpse_callback(struct ldlm_lock *lock, void *reqp)
153 struct ptlrpc_request *req = reqp;
154 struct inode *inode = llu_inode_from_lock(lock);
155 struct llu_inode_info *lli;
161 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
162 lli = llu_i2info(inode);
164 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
165 if (lli->lli_smd == NULL)
166 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
168 /* First, find out which stripe index this lock corresponds to. */
169 if (lli->lli_smd->lsm_stripe_count > 1)
170 stripe = llu_lock_to_stripe_offset(inode, lock);
172 req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
173 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
175 rc = req_capsule_server_pack(&req->rq_pill);
177 CERROR("failed pack reply: %d\n", rc);
181 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
182 lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms;
184 LDLM_DEBUG(lock, "i_size: "LPU64" -> stripe number %u -> kms "LPU64,
185 (__u64)llu_i2stat(inode)->st_size, stripe,lvb->lvb_size);
189 /* These errors are normal races, so we don't want to fill the console
190 * with messages by calling ptlrpc_error() */
191 if (rc == -ELDLM_NO_LOCK_DATA)
192 lustre_pack_reply(req, 1, NULL, NULL);
198 int llu_merge_lvb(struct inode *inode)
200 struct llu_inode_info *lli = llu_i2info(inode);
201 struct llu_sb_info *sbi = llu_i2sbi(inode);
202 struct intnl_stat *st = llu_i2stat(inode);
207 lov_stripe_lock(lli->lli_smd);
208 inode_init_lvb(inode, &lvb);
209 /* merge timestamps the most resently obtained from mds with
210 timestamps obtained from osts */
211 lvb.lvb_atime = lli->lli_lvb.lvb_atime;
212 lvb.lvb_mtime = lli->lli_lvb.lvb_mtime;
213 lvb.lvb_ctime = lli->lli_lvb.lvb_ctime;
214 rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
215 st->st_size = lvb.lvb_size;
216 st->st_blocks = lvb.lvb_blocks;
217 /* handle st_blocks overflow gracefully */
218 if (st->st_blocks < lvb.lvb_blocks)
219 st->st_blocks = ~0UL;
220 st->st_mtime = lvb.lvb_mtime;
221 st->st_atime = lvb.lvb_atime;
222 st->st_ctime = lvb.lvb_ctime;
223 lov_stripe_unlock(lli->lli_smd);
228 int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
229 struct lov_stripe_md *lsm, int mode,
230 ldlm_policy_data_t *policy, struct lustre_handle *lockh,
233 struct llu_sb_info *sbi = llu_i2sbi(inode);
234 struct intnl_stat *st = llu_i2stat(inode);
235 struct ldlm_enqueue_info einfo = { 0 };
236 struct obd_info oinfo = { { { 0 } } };
241 LASSERT(!lustre_handle_is_used(lockh));
242 CLASSERT(ELDLM_OK == 0);
244 /* XXX phil: can we do this? won't it screw the file size up? */
245 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
246 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
249 CDEBUG(D_DLMTRACE, "Locking inode "LPU64", start "LPU64" end "LPU64"\n",
250 (__u64)st->st_ino, policy->l_extent.start,
251 policy->l_extent.end);
253 einfo.ei_type = LDLM_EXTENT;
254 einfo.ei_mode = mode;
255 einfo.ei_cb_bl = llu_extent_lock_cancel_cb;
256 einfo.ei_cb_cp = ldlm_completion_ast;
257 einfo.ei_cb_gl = llu_glimpse_callback;
258 einfo.ei_cbdata = inode;
260 oinfo.oi_policy = *policy;
261 oinfo.oi_lockh = lockh;
263 oinfo.oi_flags = ast_flags;
265 rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL);
266 *policy = oinfo.oi_policy;
270 inode_init_lvb(inode, &lvb);
271 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1);
272 if (policy->l_extent.start == 0 &&
273 policy->l_extent.end == OBD_OBJECT_EOF)
274 st->st_size = lvb.lvb_size;
277 st->st_mtime = lvb.lvb_mtime;
278 st->st_atime = lvb.lvb_atime;
279 st->st_ctime = lvb.lvb_ctime;
285 int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode,
286 struct lov_stripe_md *lsm, int mode,
287 struct lustre_handle *lockh)
289 struct llu_sb_info *sbi = llu_i2sbi(inode);
293 CLASSERT(ELDLM_OK == 0);
295 /* XXX phil: can we do this? won't it screw the file size up? */
296 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
297 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
300 rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh);
306 ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen,
307 _SYSIO_OFF_T pos, ssize_t len,
310 struct llu_io_session *session = (struct llu_io_session *) private;
311 struct inode *inode = session->lis_inode;
312 struct llu_inode_info *lli = llu_i2info(inode);
320 /* in a large iov read/write we'll be repeatedly called.
321 * so give a chance to answer cancel ast here
323 liblustre_wait_event(0);
325 if (len == 0 || iovlen == 0)
328 if (pos + len > lli->lli_maxbytes)
331 env = cl_env_get(&refcheck);
333 RETURN(PTR_ERR(env));
335 /* Do NOT call "ccc_env_thread_io()" again to prevent reinitializing */
336 io = &ccc_env_info(env)->cti_io;
337 if (cl_io_rw_init(env, io, session->lis_cmd == OBD_BRW_WRITE?CIT_WRITE:
341 sio = slp_env_io(env);
342 cio = ccc_env_io(env);
343 /* XXX this is not right: cio->cui_iov can be modified. */
344 cio->cui_iov = (struct iovec *)iovec;
345 cio->cui_nrsegs = iovlen;
346 cio->cui_tot_nrsegs = iovlen;
347 sio->sio_session = session;
348 err = cl_io_loop(env, io);
354 cl_env_put(env, &refcheck);
363 struct llu_io_session *get_io_session(struct inode *ino, int ngroups, int cmd)
365 struct llu_io_session *session;
367 OBD_ALLOC_PTR(session);
372 session->lis_inode = ino;
373 session->lis_max_groups = ngroups;
374 session->lis_cmd = cmd;
378 static void put_io_session(struct llu_io_session *session)
380 I_RELE(session->lis_inode);
381 OBD_FREE_PTR(session);
384 static int llu_file_rwx(struct inode *ino,
388 struct llu_io_session *session;
390 int cmd = read ? OBD_BRW_READ : OBD_BRW_WRITE;
393 LASSERT(ioctx->ioctx_xtvlen >= 0);
394 LASSERT(ioctx->ioctx_iovlen >= 0);
396 liblustre_wait_event(0);
398 if (!ioctx->ioctx_xtvlen)
401 /* XXX consider other types later */
402 if (S_ISDIR(llu_i2stat(ino)->st_mode))
404 if (!S_ISREG(llu_i2stat(ino)->st_mode))
407 session = get_io_session(ino, ioctx->ioctx_xtvlen * 2, cmd);
411 cc = _sysio_enumerate_extents(ioctx->ioctx_xtv, ioctx->ioctx_xtvlen,
412 ioctx->ioctx_iov, ioctx->ioctx_iovlen,
413 llu_file_prwv, session);
416 LASSERT(!ioctx->ioctx_cc);
417 ioctx->ioctx_private = session;
420 put_io_session(session);
423 liblustre_wait_event(0);
427 void llu_io_init(struct cl_io *io, struct inode *inode, int write)
429 struct llu_inode_info *lli = llu_i2info(inode);
431 io->u.ci_rw.crw_nonblock = lli->lli_open_flags & O_NONBLOCK;
433 io->u.ci_wr.wr_append = lli->lli_open_flags & O_APPEND;
434 io->ci_obj = llu_i2info(inode)->lli_clob;
436 if ((lli->lli_open_flags & O_APPEND) && write)
437 io->ci_lockreq = CILR_MANDATORY;
439 io->ci_lockreq = CILR_NEVER;
442 int llu_iop_read(struct inode *ino,
445 struct intnl_stat *st = llu_i2stat(ino);
452 st->st_atime = CFS_CURRENT_TIME;
454 env = cl_env_get(&refcheck);
456 RETURN(PTR_ERR(env));
458 io = ccc_env_thread_io(env);
459 llu_io_init(io, ino, 0);
461 ret = llu_file_rwx(ino, ioctx, 1);
463 cl_env_put(env, &refcheck);
467 int llu_iop_write(struct inode *ino,
470 struct intnl_stat *st = llu_i2stat(ino);
476 st->st_mtime = st->st_ctime = CFS_CURRENT_TIME;
478 env = cl_env_get(&refcheck);
480 RETURN(PTR_ERR(env));
482 io = ccc_env_thread_io(env);
483 llu_io_init(io, ino, 1);
485 ret = llu_file_rwx(ino, ioctx, 0);
486 cl_env_put(env, &refcheck);
490 int llu_iop_iodone(struct ioctx *ioctx)
492 struct llu_io_session *session;
498 liblustre_wait_event(0);
500 env = cl_env_get(&refcheck);
502 RETURN(PTR_ERR(env));
504 io = &ccc_env_info(env)->cti_io;
506 cl_env_put(env, &refcheck);
507 session = (struct llu_io_session *) ioctx->ioctx_private;
509 LASSERT(!IS_ERR(session));
511 if (session->lis_rc == 0) {
512 ioctx->ioctx_cc = session->lis_rwcount;
514 LASSERT(session->lis_rc < 0);
515 ioctx->ioctx_cc = -1;
516 ioctx->ioctx_errno = -session->lis_rc;
519 put_io_session(session);
520 ioctx->ioctx_private = NULL;
521 liblustre_wait_event(0);