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.
32 * Copyright (c) 2011, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/liblustre/rw.c
40 * Lustre Light block IO
43 #define DEBUG_SUBSYSTEM S_LLITE
49 #include <sys/types.h>
51 #include <sys/queue.h>
55 #include "llite_lib.h"
57 typedef ssize_t llu_file_piov_t(const struct iovec *iovec, int iovlen,
58 _SYSIO_OFF_T pos, ssize_t len,
61 size_t llap_cookie_size;
63 static int llu_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock)
65 struct llu_inode_info *lli = llu_i2info(inode);
66 struct lov_stripe_md *lsm = lli->lli_smd;
67 struct obd_export *exp = llu_i2obdexp(inode);
70 struct ldlm_lock *lock;
71 } key = { .name = KEY_LOCK_TO_STRIPE, .lock = lock };
72 __u32 stripe, vallen = sizeof(stripe);
76 if (lsm->lsm_stripe_count == 1)
79 /* get our offset in the lov */
80 rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe, lsm);
82 CERROR("obd_get_info: rc = %d\n", rc);
85 LASSERT(stripe < lsm->lsm_stripe_count);
89 int llu_extent_lock_cancel_cb(struct ldlm_lock *lock,
90 struct ldlm_lock_desc *new, void *data,
93 struct lustre_handle lockh = { 0 };
97 if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
98 LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
103 case LDLM_CB_BLOCKING:
104 ldlm_lock2handle(lock, &lockh);
105 rc = ldlm_cli_cancel(&lockh);
107 CERROR("ldlm_cli_cancel failed: %d\n", rc);
109 case LDLM_CB_CANCELING: {
111 struct llu_inode_info *lli;
112 struct lov_stripe_md *lsm;
116 /* This lock wasn't granted, don't try to evict pages */
117 if (lock->l_req_mode != lock->l_granted_mode)
120 inode = llu_inode_from_lock(lock);
123 lli= llu_i2info(inode);
130 stripe = llu_lock_to_stripe_offset(inode, lock);
131 lock_res_and_lock(lock);
132 kms = ldlm_extent_shift_kms(lock,
133 lsm->lsm_oinfo[stripe]->loi_kms);
134 unlock_res_and_lock(lock);
135 if (lsm->lsm_oinfo[stripe]->loi_kms != kms)
136 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
137 lsm->lsm_oinfo[stripe]->loi_kms, kms);
138 loi_kms_set(lsm->lsm_oinfo[stripe], kms);
150 static int llu_glimpse_callback(struct ldlm_lock *lock, void *reqp)
152 struct ptlrpc_request *req = reqp;
153 struct inode *inode = llu_inode_from_lock(lock);
154 struct llu_inode_info *lli;
160 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
161 lli = llu_i2info(inode);
163 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
164 if (lli->lli_smd == NULL)
165 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
167 /* First, find out which stripe index this lock corresponds to. */
168 if (lli->lli_smd->lsm_stripe_count > 1)
169 stripe = llu_lock_to_stripe_offset(inode, lock);
171 req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK);
172 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
174 rc = req_capsule_server_pack(&req->rq_pill);
176 CERROR("failed pack reply: %d\n", rc);
180 lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB);
181 lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms;
183 LDLM_DEBUG(lock, "i_size: "LPU64" -> stripe number %u -> kms "LPU64,
184 (__u64)llu_i2stat(inode)->st_size, stripe,lvb->lvb_size);
188 /* These errors are normal races, so we don't want to fill the console
189 * with messages by calling ptlrpc_error() */
190 if (rc == -ELDLM_NO_LOCK_DATA)
191 lustre_pack_reply(req, 1, NULL, NULL);
197 int llu_merge_lvb(struct inode *inode)
199 struct llu_inode_info *lli = llu_i2info(inode);
200 struct llu_sb_info *sbi = llu_i2sbi(inode);
201 struct intnl_stat *st = llu_i2stat(inode);
206 lov_stripe_lock(lli->lli_smd);
207 inode_init_lvb(inode, &lvb);
208 /* merge timestamps the most resently obtained from mds with
209 timestamps obtained from osts */
210 lvb.lvb_atime = lli->lli_lvb.lvb_atime;
211 lvb.lvb_mtime = lli->lli_lvb.lvb_mtime;
212 lvb.lvb_ctime = lli->lli_lvb.lvb_ctime;
213 rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
214 st->st_size = lvb.lvb_size;
215 st->st_blocks = lvb.lvb_blocks;
216 /* handle st_blocks overflow gracefully */
217 if (st->st_blocks < lvb.lvb_blocks)
218 st->st_blocks = ~0UL;
219 st->st_mtime = lvb.lvb_mtime;
220 st->st_atime = lvb.lvb_atime;
221 st->st_ctime = lvb.lvb_ctime;
222 lov_stripe_unlock(lli->lli_smd);
227 int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
228 struct lov_stripe_md *lsm, int mode,
229 ldlm_policy_data_t *policy, struct lustre_handle *lockh,
232 struct llu_sb_info *sbi = llu_i2sbi(inode);
233 struct intnl_stat *st = llu_i2stat(inode);
234 struct ldlm_enqueue_info einfo = { 0 };
235 struct obd_info oinfo = { { { 0 } } };
240 LASSERT(!lustre_handle_is_used(lockh));
241 CLASSERT(ELDLM_OK == 0);
243 /* XXX phil: can we do this? won't it screw the file size up? */
244 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
245 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
248 CDEBUG(D_DLMTRACE, "Locking inode "LPU64", start "LPU64" end "LPU64"\n",
249 (__u64)st->st_ino, policy->l_extent.start,
250 policy->l_extent.end);
252 einfo.ei_type = LDLM_EXTENT;
253 einfo.ei_mode = mode;
254 einfo.ei_cb_bl = llu_extent_lock_cancel_cb;
255 einfo.ei_cb_cp = ldlm_completion_ast;
256 einfo.ei_cb_gl = llu_glimpse_callback;
257 einfo.ei_cbdata = inode;
259 oinfo.oi_policy = *policy;
260 oinfo.oi_lockh = lockh;
262 oinfo.oi_flags = ast_flags;
264 rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL);
265 *policy = oinfo.oi_policy;
269 inode_init_lvb(inode, &lvb);
270 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1);
271 if (policy->l_extent.start == 0 &&
272 policy->l_extent.end == OBD_OBJECT_EOF)
273 st->st_size = lvb.lvb_size;
276 st->st_mtime = lvb.lvb_mtime;
277 st->st_atime = lvb.lvb_atime;
278 st->st_ctime = lvb.lvb_ctime;
284 int llu_extent_unlock(struct ll_file_data *fd, struct inode *inode,
285 struct lov_stripe_md *lsm, int mode,
286 struct lustre_handle *lockh)
288 struct llu_sb_info *sbi = llu_i2sbi(inode);
292 CLASSERT(ELDLM_OK == 0);
294 /* XXX phil: can we do this? won't it screw the file size up? */
295 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
296 (sbi->ll_flags & LL_SBI_NOLCK) || mode == LCK_NL)
299 rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh);
305 ssize_t llu_file_prwv(const struct iovec *iovec, int iovlen,
306 _SYSIO_OFF_T pos, ssize_t len,
309 struct llu_io_session *session = (struct llu_io_session *) private;
310 struct inode *inode = session->lis_inode;
311 struct llu_inode_info *lli = llu_i2info(inode);
319 /* in a large iov read/write we'll be repeatedly called.
320 * so give a chance to answer cancel ast here
322 liblustre_wait_event(0);
324 if (len == 0 || iovlen == 0)
327 if (pos + len > lli->lli_maxbytes)
330 env = cl_env_get(&refcheck);
332 RETURN(PTR_ERR(env));
334 /* Do NOT call "ccc_env_thread_io()" again to prevent reinitializing */
335 io = &ccc_env_info(env)->cti_io;
336 if (cl_io_rw_init(env, io, session->lis_cmd == OBD_BRW_WRITE?CIT_WRITE:
340 sio = slp_env_io(env);
341 cio = ccc_env_io(env);
342 /* XXX this is not right: cio->cui_iov can be modified. */
343 cio->cui_iov = (struct iovec *)iovec;
344 cio->cui_nrsegs = iovlen;
345 cio->cui_tot_nrsegs = iovlen;
346 sio->sio_session = session;
347 err = cl_io_loop(env, io);
353 cl_env_put(env, &refcheck);
362 struct llu_io_session *get_io_session(struct inode *ino, int ngroups, int cmd)
364 struct llu_io_session *session;
366 OBD_ALLOC_PTR(session);
371 session->lis_inode = ino;
372 session->lis_max_groups = ngroups;
373 session->lis_cmd = cmd;
377 static void put_io_session(struct llu_io_session *session)
379 I_RELE(session->lis_inode);
380 OBD_FREE_PTR(session);
383 static int llu_file_rwx(struct inode *ino,
387 struct llu_io_session *session;
389 int cmd = read ? OBD_BRW_READ : OBD_BRW_WRITE;
392 LASSERT(ioctx->ioctx_xtvlen >= 0);
393 LASSERT(ioctx->ioctx_iovlen >= 0);
395 liblustre_wait_event(0);
397 if (!ioctx->ioctx_xtvlen)
400 /* XXX consider other types later */
401 if (S_ISDIR(llu_i2stat(ino)->st_mode))
403 if (!S_ISREG(llu_i2stat(ino)->st_mode))
406 session = get_io_session(ino, ioctx->ioctx_xtvlen * 2, cmd);
410 cc = _sysio_enumerate_extents(ioctx->ioctx_xtv, ioctx->ioctx_xtvlen,
411 ioctx->ioctx_iov, ioctx->ioctx_iovlen,
412 llu_file_prwv, session);
415 LASSERT(!ioctx->ioctx_cc);
416 ioctx->ioctx_private = session;
419 put_io_session(session);
422 liblustre_wait_event(0);
426 void llu_io_init(struct cl_io *io, struct inode *inode, int write)
428 struct llu_inode_info *lli = llu_i2info(inode);
430 io->u.ci_rw.crw_nonblock = lli->lli_open_flags & O_NONBLOCK;
432 io->u.ci_wr.wr_append = lli->lli_open_flags & O_APPEND;
433 io->ci_obj = llu_i2info(inode)->lli_clob;
435 if ((lli->lli_open_flags & O_APPEND) && write)
436 io->ci_lockreq = CILR_MANDATORY;
438 io->ci_lockreq = CILR_NEVER;
441 int llu_iop_read(struct inode *ino,
444 struct intnl_stat *st = llu_i2stat(ino);
451 st->st_atime = CFS_CURRENT_TIME;
453 env = cl_env_get(&refcheck);
455 RETURN(PTR_ERR(env));
457 io = ccc_env_thread_io(env);
458 llu_io_init(io, ino, 0);
460 ret = llu_file_rwx(ino, ioctx, 1);
462 cl_env_put(env, &refcheck);
466 int llu_iop_write(struct inode *ino,
469 struct intnl_stat *st = llu_i2stat(ino);
475 st->st_mtime = st->st_ctime = CFS_CURRENT_TIME;
477 env = cl_env_get(&refcheck);
479 RETURN(PTR_ERR(env));
481 io = ccc_env_thread_io(env);
482 llu_io_init(io, ino, 1);
484 ret = llu_file_rwx(ino, ioctx, 0);
485 cl_env_put(env, &refcheck);
489 int llu_iop_iodone(struct ioctx *ioctx)
491 struct llu_io_session *session;
497 liblustre_wait_event(0);
499 env = cl_env_get(&refcheck);
501 RETURN(PTR_ERR(env));
503 io = &ccc_env_info(env)->cti_io;
505 cl_env_put(env, &refcheck);
506 session = (struct llu_io_session *) ioctx->ioctx_private;
508 LASSERT(!IS_ERR(session));
510 if (session->lis_rc == 0) {
511 ioctx->ioctx_cc = session->lis_rwcount;
513 LASSERT(session->lis_rc < 0);
514 ioctx->ioctx_cc = -1;
515 ioctx->ioctx_errno = -session->lis_rc;
518 put_io_session(session);
519 ioctx->ioctx_private = NULL;
520 liblustre_wait_event(0);