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 2008 Sun Microsystems, Inc. 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>
66 #include "llite_lib.h"
68 typedef ssize_t llu_file_piov_t(const struct iovec *iovec, int iovlen,
69 _SYSIO_OFF_T pos, ssize_t len,
72 size_t llap_cookie_size;
74 static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock)
76 struct llu_inode_info *lli = llu_i2info(inode);
77 struct lov_stripe_md *lsm = lli->lli_smd;
78 struct obd_export *exp = llu_i2obdexp(inode);
81 struct ldlm_lock *lock;
82 } key = { .name = KEY_LOCK_TO_STRIPE, .lock = lock };
83 __u32 stripe, vallen = sizeof(stripe);
87 if (lsm->lsm_stripe_count == 1)
90 /* get our offset in the lov */
91 rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe, lsm);
93 CERROR("obd_get_info: rc = %d\n", rc);
96 LASSERT(stripe < lsm->lsm_stripe_count);
100 int llu_extent_lock_cancel_cb(struct ldlm_lock *lock,
101 struct ldlm_lock_desc *new, void *data,
104 struct lustre_handle lockh = { 0 };
108 if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
109 LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
114 case LDLM_CB_BLOCKING:
115 ldlm_lock2handle(lock, &lockh);
116 rc = ldlm_cli_cancel(&lockh);
118 CERROR("ldlm_cli_cancel failed: %d\n", rc);
120 case LDLM_CB_CANCELING: {
122 struct llu_inode_info *lli;
123 struct lov_stripe_md *lsm;
127 /* This lock wasn't granted, don't try to evict pages */
128 if (lock->l_req_mode != lock->l_granted_mode)
131 inode = llu_inode_from_lock(lock);
134 lli= llu_i2info(inode);
141 stripe = llu_lock_to_stripe_offset(inode, lock);
142 lock_res_and_lock(lock);
143 kms = ldlm_extent_shift_kms(lock,
144 lsm->lsm_oinfo[stripe]->loi_kms);
145 unlock_res_and_lock(lock);
146 if (lsm->lsm_oinfo[stripe]->loi_kms != kms)
147 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
148 lsm->lsm_oinfo[stripe]->loi_kms, kms);
149 loi_kms_set(lsm->lsm_oinfo[stripe], kms);
161 static int llu_glimpse_callback(struct ldlm_lock *lock, void *reqp)
163 struct ptlrpc_request *req = reqp;
164 struct inode *inode = llu_inode_from_lock(lock);
165 struct llu_inode_info *lli;
171 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
172 lli = llu_i2info(inode);
174 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
175 if (lli->lli_smd == NULL)
176 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
178 /* First, find out which stripe index this lock corresponds to. */
179 if (lli->lli_smd->lsm_stripe_count > 1)
180 stripe = llu_lock_to_stripe_offset(inode, lock);
182 req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
183 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
185 rc = req_capsule_server_pack(&req->rq_pill);
187 CERROR("failed pack reply: %d\n", rc);
191 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
192 lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms;
194 LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64,
195 (__u64)llu_i2stat(inode)->st_size, stripe,lvb->lvb_size);
199 /* These errors are normal races, so we don't want to fill the console
200 * with messages by calling ptlrpc_error() */
201 if (rc == -ELDLM_NO_LOCK_DATA)
202 lustre_pack_reply(req, 1, NULL, NULL);
208 int llu_merge_lvb(struct inode *inode)
210 struct llu_inode_info *lli = llu_i2info(inode);
211 struct llu_sb_info *sbi = llu_i2sbi(inode);
212 struct intnl_stat *st = llu_i2stat(inode);
217 inode_init_lvb(inode, &lvb);
218 rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
219 st->st_size = lvb.lvb_size;
220 st->st_blocks = lvb.lvb_blocks;
221 /* handle st_blocks overflow gracefully */
222 if (st->st_blocks < lvb.lvb_blocks)
223 st->st_blocks = ~0UL;
224 st->st_mtime = lvb.lvb_mtime;
225 st->st_atime = lvb.lvb_atime;
226 st->st_ctime = lvb.lvb_ctime;
231 int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
232 struct lov_stripe_md *lsm, int mode,
233 ldlm_policy_data_t *policy, struct lustre_handle *lockh,
236 struct llu_sb_info *sbi = llu_i2sbi(inode);
237 struct intnl_stat *st = llu_i2stat(inode);
238 struct ldlm_enqueue_info einfo = { 0 };
239 struct obd_info oinfo = { { { 0 } } };
244 LASSERT(!lustre_handle_is_used(lockh));
245 CLASSERT(ELDLM_OK == 0);
247 /* XXX phil: can we do this? won't it screw the file size up? */
248 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
249 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
252 CDEBUG(D_DLMTRACE, "Locking inode %llu, start "LPU64" end "LPU64"\n",
253 (__u64)st->st_ino, policy->l_extent.start,
254 policy->l_extent.end);
256 einfo.ei_type = LDLM_EXTENT;
257 einfo.ei_mode = mode;
258 einfo.ei_cb_bl = llu_extent_lock_cancel_cb;
259 einfo.ei_cb_cp = ldlm_completion_ast;
260 einfo.ei_cb_gl = llu_glimpse_callback;
261 einfo.ei_cbdata = inode;
263 oinfo.oi_policy = *policy;
264 oinfo.oi_lockh = lockh;
266 oinfo.oi_flags = ast_flags;
268 rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL);
269 *policy = oinfo.oi_policy;
273 inode_init_lvb(inode, &lvb);
274 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1);
275 if (policy->l_extent.start == 0 &&
276 policy->l_extent.end == OBD_OBJECT_EOF)
277 st->st_size = lvb.lvb_size;
280 st->st_mtime = lvb.lvb_mtime;
281 st->st_atime = lvb.lvb_atime;
282 st->st_ctime = lvb.lvb_ctime;
288 int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode,
289 struct lov_stripe_md *lsm, int mode,
290 struct lustre_handle *lockh)
292 struct llu_sb_info *sbi = llu_i2sbi(inode);
296 CLASSERT(ELDLM_OK == 0);
298 /* XXX phil: can we do this? won't it screw the file size up? */
299 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
300 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
303 rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh);
309 ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen,
310 _SYSIO_OFF_T pos, ssize_t len,
313 struct llu_io_session *session = (struct llu_io_session *) private;
314 struct inode *inode = session->lis_inode;
315 struct llu_inode_info *lli = llu_i2info(inode);
323 /* in a large iov read/write we'll be repeatedly called.
324 * so give a chance to answer cancel ast here
326 liblustre_wait_event(0);
328 if (len == 0 || iovlen == 0)
331 if (pos + len > lli->lli_maxbytes)
334 env = cl_env_get(&refcheck);
336 RETURN(PTR_ERR(env));
338 io = &ccc_env_info(env)->cti_io;
340 if (cl_io_rw_init(env, io, session->lis_cmd == OBD_BRW_WRITE?CIT_WRITE:
344 sio = slp_env_io(env);
345 cio = ccc_env_io(env);
346 /* XXX this is not right: cio->cui_iov can be modified. */
347 cio->cui_iov = (struct iovec *)iovec;
348 cio->cui_nrsegs = iovlen;
349 sio->sio_session = session;
350 err = cl_io_loop(env, io);
356 cl_env_put(env, &refcheck);
365 struct llu_io_session *get_io_session(struct inode *ino, int ngroups, int cmd)
367 struct llu_io_session *session;
369 OBD_ALLOC(session, LLU_IO_SESSION_SIZE(ngroups));
374 session->lis_inode = ino;
375 session->lis_max_groups = ngroups;
376 session->lis_cmd = cmd;
380 static void put_io_session(struct llu_io_session *session)
384 for (i = 0; i < session->lis_ngroups; i++) {
385 if (session->lis_groups[i]) {
386 put_io_group(session->lis_groups[i]);
387 session->lis_groups[i] = NULL;
391 I_RELE(session->lis_inode);
392 OBD_FREE(session, LLU_IO_SESSION_SIZE(session->lis_max_groups));
395 static int llu_file_rwx(struct inode *ino,
399 struct llu_io_session *session;
401 int cmd = read ? OBD_BRW_READ : OBD_BRW_WRITE;
404 LASSERT(ioctx->ioctx_xtvlen >= 0);
405 LASSERT(ioctx->ioctx_iovlen >= 0);
407 liblustre_wait_event(0);
409 if (!ioctx->ioctx_xtvlen)
412 /* XXX consider other types later */
413 if (S_ISDIR(llu_i2stat(ino)->st_mode))
415 if (!S_ISREG(llu_i2stat(ino)->st_mode))
418 session = get_io_session(ino, ioctx->ioctx_xtvlen * 2, cmd);
422 cc = _sysio_enumerate_extents(ioctx->ioctx_xtv, ioctx->ioctx_xtvlen,
423 ioctx->ioctx_iov, ioctx->ioctx_iovlen,
424 llu_file_prwv, session);
427 LASSERT(!ioctx->ioctx_cc);
428 ioctx->ioctx_private = session;
431 put_io_session(session);
434 liblustre_wait_event(0);
438 void llu_io_init(struct cl_io *io, struct inode *inode, int write)
440 struct llu_inode_info *lli = llu_i2info(inode);
442 memset(io, 0, sizeof *io);
444 io->u.ci_rw.crw_nonblock = lli->lli_open_flags & O_NONBLOCK;
446 io->u.ci_wr.wr_append = lli->lli_open_flags & O_APPEND;
447 io->ci_obj = llu_i2info(inode)->lli_clob;
449 if (lli->lli_open_flags & O_APPEND)
450 io->ci_lockreq = CILR_MANDATORY;
452 io->ci_lockreq = CILR_NEVER;
456 int llu_iop_read(struct inode *ino,
459 struct intnl_stat *st = llu_i2stat(ino);
466 st->st_atime = CURRENT_TIME;
468 env = cl_env_get(&refcheck);
470 RETURN(PTR_ERR(env));
472 io = &ccc_env_info(env)->cti_io;
473 llu_io_init(io, ino, 0);
475 ret = llu_file_rwx(ino, ioctx, 1);
477 cl_env_put(env, &refcheck);
481 int llu_iop_write(struct inode *ino,
484 struct intnl_stat *st = llu_i2stat(ino);
490 st->st_mtime = st->st_ctime = CURRENT_TIME;
492 env = cl_env_get(&refcheck);
494 RETURN(PTR_ERR(env));
496 io = &ccc_env_info(env)->cti_io;
497 llu_io_init(io, ino, 1);
499 ret = llu_file_rwx(ino, ioctx, 0);
500 cl_env_put(env, &refcheck);
504 int llu_iop_iodone(struct ioctx *ioctx)
506 struct llu_io_session *session;
507 struct llu_io_group *group;
514 liblustre_wait_event(0);
516 env = cl_env_get(&refcheck);
518 RETURN(PTR_ERR(env));
520 io = &ccc_env_info(env)->cti_io;
522 cl_env_put(env, &refcheck);
523 session = (struct llu_io_session *) ioctx->ioctx_private;
525 LASSERT(!IS_ERR(session));
527 for (i = 0; i < session->lis_ngroups; i++) {
528 group = session->lis_groups[i];
533 ioctx->ioctx_cc += group->lig_rwcount;
535 session->lis_groups[i] = NULL;
541 ioctx->ioctx_cc = -1;
542 ioctx->ioctx_errno = -rc;
545 put_io_session(session);
546 ioctx->ioctx_private = NULL;
547 liblustre_wait_event(0);