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/file.c
38 * Lustre Light file operations
41 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <sys/types.h>
49 #include <sys/queue.h>
63 #include "llite_lib.h"
65 /* Pack the required supplementary groups into the supplied groups array.
66 * If we don't need to use the groups from the target inode(s) then we
67 * instead pack one or more groups from the user's supplementary group
68 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
69 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
72 LASSERT(suppgids != NULL);
74 if (cfs_curproc_is_in_groups(i1->i_stbuf.st_gid))
75 suppgids[0] = i1->i_stbuf.st_gid;
80 if (cfs_curproc_is_in_groups(i2->i_stbuf.st_gid))
81 suppgids[1] = i2->i_stbuf.st_gid;
89 void llu_prep_md_op_data(struct md_op_data *op_data, struct inode *i1,
90 struct inode *i2, const char *name, int namelen,
93 LASSERT(i1 != NULL || i2 != NULL);
97 ll_i2gids(op_data->op_suppgids, i1, i2);
98 op_data->op_fid1 = *ll_inode2fid(i1);
100 ll_i2gids(op_data->op_suppgids, i2, i1);
101 op_data->op_fid1 = *ll_inode2fid(i2);
105 op_data->op_fid2 = *ll_inode2fid(i2);
107 fid_zero(&op_data->op_fid2);
109 op_data->op_opc = opc;
110 op_data->op_name = name;
111 op_data->op_mode = mode;
112 op_data->op_namelen = namelen;
113 op_data->op_mod_time = CFS_CURRENT_TIME;
114 op_data->op_data = NULL;
117 void llu_finish_md_op_data(struct md_op_data *op_data)
119 OBD_FREE_PTR(op_data);
122 void obdo_refresh_inode(struct inode *dst,
126 struct intnl_stat *st = llu_i2stat(dst);
127 valid &= src->o_valid;
129 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
130 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
132 src->o_valid, LTIME_S(st->st_mtime),
133 LTIME_S(st->st_ctime),
134 (long)src->o_mtime, (long)src->o_ctime);
136 if (valid & OBD_MD_FLATIME && src->o_atime > LTIME_S(st->st_atime))
137 LTIME_S(st->st_atime) = src->o_atime;
138 if (valid & OBD_MD_FLMTIME && src->o_mtime > LTIME_S(st->st_mtime))
139 LTIME_S(st->st_mtime) = src->o_mtime;
140 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
141 LTIME_S(st->st_ctime) = src->o_ctime;
142 if (valid & OBD_MD_FLSIZE && src->o_size > st->st_size)
143 st->st_size = src->o_size;
144 /* optimum IO size */
145 if (valid & OBD_MD_FLBLKSZ)
146 st->st_blksize = src->o_blksize;
147 /* allocation of space */
148 if (valid & OBD_MD_FLBLOCKS && src->o_blocks > st->st_blocks)
149 st->st_blocks = src->o_blocks;
153 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
154 * not believe attributes if a few ioepoch holders exist. Attributes for
155 * previous ioepoch if new one is opened are also skipped by MDS.
157 void llu_ioepoch_open(struct llu_inode_info *lli, __u64 ioepoch)
159 if (ioepoch && lli->lli_ioepoch != ioepoch) {
160 lli->lli_ioepoch = ioepoch;
161 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for truncate\n",
162 ioepoch, PFID(&lli->lli_fid));
166 int llu_local_open(struct llu_inode_info *lli, struct lookup_intent *it)
168 struct ptlrpc_request *req = it->d.lustre.it_data;
169 struct ll_file_data *fd;
170 struct mdt_body *body;
173 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
174 LASSERT(body != NULL);
176 /* already opened? */
177 if (lli->lli_open_count++)
180 LASSERT(!lli->lli_file_data);
182 OBD_ALLOC(fd, sizeof(*fd));
183 /* We can't handle this well without reorganizing ll_file_open and
184 * ll_md_close, so don't even try right now. */
187 memcpy(&fd->fd_mds_och.och_fh, &body->handle, sizeof(body->handle));
188 fd->fd_mds_och.och_magic = OBD_CLIENT_HANDLE_MAGIC;
189 fd->fd_mds_och.och_fid = lli->lli_fid;
190 lli->lli_file_data = fd;
191 llu_ioepoch_open(lli, body->ioepoch);
192 md_set_open_replay_data(lli->lli_sbi->ll_md_exp,
193 &fd->fd_mds_och, it->d.lustre.it_data);
198 int llu_iop_open(struct pnode *pnode, int flags, mode_t mode)
200 struct inode *inode = pnode->p_base->pb_ino;
201 struct llu_inode_info *lli = llu_i2info(inode);
202 struct intnl_stat *st = llu_i2stat(inode);
203 struct ll_file_data *fd;
204 struct ptlrpc_request *request;
205 struct lookup_intent *it;
206 struct lov_stripe_md *lsm;
210 liblustre_wait_event(0);
212 /* don't do anything for '/' */
213 if (llu_is_root_inode(inode))
216 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
217 LL_GET_INTENT(inode, it);
219 if (!it->d.lustre.it_disposition) {
223 rc = it_open_error(DISP_OPEN_OPEN, it);
225 GOTO(out_release, rc);
227 rc = llu_local_open(lli, it);
231 if (!S_ISREG(st->st_mode))
232 GOTO(out_release, rc = 0);
234 fd = lli->lli_file_data;
238 flags &= ~O_LOV_DELAY_CREATE;
239 /*XXX: open_flags are overwritten and the previous ones are lost */
240 lli->lli_open_flags = flags & ~(O_CREAT | O_EXCL | O_TRUNC);
243 request = it->d.lustre.it_data;
244 ptlrpc_req_finished(request);
246 it->it_op_release(it);
247 OBD_FREE(it, sizeof(*it));
249 /* libsysio hasn't done anything for O_TRUNC. here we
250 * simply simulate it as open(...); truncate(...); */
251 if (rc == 0 && (flags & O_TRUNC) && S_ISREG(st->st_mode)) {
254 memset(&attr, 0, sizeof(attr));
256 attr.ia_valid |= ATTR_SIZE | ATTR_RAW;
257 rc = llu_setattr_raw(inode, &attr);
259 CERROR("error %d truncate in open()\n", rc);
262 liblustre_wait_event(0);
266 int llu_objects_destroy(struct ptlrpc_request *req, struct inode *dir)
268 struct mdt_body *body;
269 struct lov_mds_md *eadata;
270 struct lov_stripe_md *lsm = NULL;
271 struct obd_trans_info oti = { 0 };
276 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
278 if (!(body->valid & OBD_MD_FLEASIZE))
281 if (body->eadatasize == 0) {
282 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
283 GOTO(out, rc = -EPROTO);
286 /* The MDS sent back the EA because we unlinked the last reference
287 * to this file. Use this EA to unlink the objects on the OST.
288 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
289 * check it is complete and sensible. */
290 eadata = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD,
293 LASSERT(eadata != NULL);
295 rc = obd_unpackmd(llu_i2obdexp(dir), &lsm, eadata,body->eadatasize);
297 CERROR("obd_unpackmd: %d\n", rc);
300 LASSERT(rc >= sizeof(*lsm));
304 GOTO(out_free_memmd, rc = -ENOMEM);
306 oa->o_id = lsm->lsm_object_id;
307 oa->o_gr = lsm->lsm_object_gr;
308 oa->o_mode = body->mode & S_IFMT;
309 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
311 if (body->valid & OBD_MD_FLCOOKIE) {
312 oa->o_valid |= OBD_MD_FLCOOKIE;
314 req_capsule_server_sized_get(&req->rq_pill,
316 sizeof(struct llog_cookie) *
317 lsm->lsm_stripe_count);
318 if (oti.oti_logcookies == NULL) {
319 oa->o_valid &= ~OBD_MD_FLCOOKIE;
320 body->valid &= ~OBD_MD_FLCOOKIE;
324 rc = obd_destroy(llu_i2obdexp(dir), oa, lsm, &oti, NULL, NULL);
327 CERROR("obd destroy objid 0x"LPX64" error %d\n",
328 lsm->lsm_object_id, rc);
330 obd_free_memmd(llu_i2obdexp(dir), &lsm);
335 /** Cliens updates SOM attributes on MDS: obd_getattr and md_setattr. */
336 int llu_som_update(struct inode *inode, struct md_op_data *op_data)
338 struct llu_inode_info *lli = llu_i2info(inode);
339 struct llu_sb_info *sbi = llu_i2sbi(inode);
340 struct obdo oa = { 0 };
345 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
346 LASSERT(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM);
348 old_flags = op_data->op_flags;
349 op_data->op_flags = MF_SOM_CHANGE;
351 /* If inode is already in another epoch, skip getattr from OSTs. */
352 if (lli->lli_ioepoch == op_data->op_ioepoch) {
353 rc = llu_inode_getattr(inode, &oa, op_data->op_ioepoch,
354 old_flags & MF_GETATTR_LOCK);
358 CDEBUG(D_INODE, "objid "LPX64" is destroyed\n",
359 lli->lli_smd->lsm_object_id);
361 CERROR("inode_getattr failed (%d): unable to "
362 "send a Size-on-MDS attribute update "
363 "for inode %llu/%lu\n", rc,
364 (long long)llu_i2stat(inode)->st_ino,
365 lli->lli_st_generation);
367 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
368 PFID(&lli->lli_fid));
371 /* Install attributes into op_data. */
372 md_from_obdo(op_data, &oa, oa.o_valid);
375 rc = llu_md_setattr(inode, op_data, NULL);
379 void llu_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
380 struct lustre_handle *fh)
382 struct llu_inode_info *lli = llu_i2info(inode);
383 struct intnl_stat *st = llu_i2stat(inode);
386 op_data->op_fid1 = lli->lli_fid;
387 op_data->op_attr.ia_atime = st->st_atime;
388 op_data->op_attr.ia_mtime = st->st_mtime;
389 op_data->op_attr.ia_ctime = st->st_ctime;
390 op_data->op_attr.ia_size = st->st_size;
391 op_data->op_attr_blocks = st->st_blocks;
392 op_data->op_attr.ia_attr_flags = lli->lli_st_flags;
393 op_data->op_ioepoch = lli->lli_ioepoch;
395 op_data->op_handle = *fh;
399 /** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
400 void llu_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
402 struct llu_inode_info *lli = llu_i2info(inode);
405 op_data->op_flags |= MF_SOM_CHANGE;
407 /* Pack Size-on-MDS attributes if we are in IO
408 * epoch and attributes are valid. */
409 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
410 if (!cl_local_size(inode))
411 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
412 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
417 static void llu_prepare_close(struct inode *inode, struct md_op_data *op_data,
418 struct ll_file_data *fd)
420 struct obd_client_handle *och = &fd->fd_mds_och;
422 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
423 ATTR_MTIME_SET | ATTR_CTIME_SET;
425 if (fd->fd_flags & FMODE_WRITE) {
426 struct llu_sb_info *sbi = llu_i2sbi(inode);
427 if (!(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM) ||
428 !S_ISREG(llu_i2stat(inode)->st_mode)) {
429 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
431 /* Inode cannot be dirty. Close the epoch. */
432 op_data->op_flags |= MF_EPOCH_CLOSE;
433 /* XXX: Send SOM attributes only if they are really
435 llu_done_writing_attr(inode, op_data);
438 llu_pack_inode2opdata(inode, op_data, &och->och_fh);
439 llu_prep_md_op_data(op_data, inode, NULL, NULL,
440 0, 0, LUSTRE_OPC_ANY);
443 int llu_md_close(struct obd_export *md_exp, struct inode *inode)
445 struct llu_inode_info *lli = llu_i2info(inode);
446 struct ll_file_data *fd = lli->lli_file_data;
447 struct ptlrpc_request *req = NULL;
448 struct obd_client_handle *och = &fd->fd_mds_och;
449 struct intnl_stat *st = llu_i2stat(inode);
450 struct md_op_data op_data = { { 0 } };
454 /* clear group lock, if present */
455 if (fd->fd_flags & LL_FILE_GROUP_LOCKED)
456 llu_put_grouplock(inode, fd->fd_grouplock.cg_gid);
458 llu_prepare_close(inode, &op_data, fd);
459 rc = md_close(md_exp, &op_data, och->och_mod, &req);
461 /* We are the last writer, so the MDS has instructed us to get
462 * the file size and any write cookies, then close again. */
463 LASSERT(lli->lli_open_flags & FMODE_WRITE);
464 rc = llu_som_update(inode, &op_data);
466 CERROR("inode %llu mdc Size-on-MDS update failed: "
467 "rc = %d\n", (long long)st->st_ino, rc);
471 CERROR("inode %llu close failed: rc %d\n",
472 (long long)st->st_ino, rc);
474 rc = llu_objects_destroy(req, inode);
476 CERROR("inode %llu ll_objects destroy: rc = %d\n",
477 (long long)st->st_ino, rc);
480 md_clear_open_replay_data(md_exp, och);
481 ptlrpc_req_finished(req);
482 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
483 lli->lli_file_data = NULL;
484 OBD_FREE(fd, sizeof(*fd));
489 int llu_file_release(struct inode *inode)
491 struct ll_file_data *fd;
492 struct llu_sb_info *sbi = llu_i2sbi(inode);
493 struct llu_inode_info *lli = llu_i2info(inode);
497 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu\n",
498 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation);
500 if (llu_is_root_inode(inode))
503 /* still opened by others? */
504 if (--lli->lli_open_count)
507 fd = lli->lli_file_data;
508 if (!fd) /* no process opened the file after an mcreate */
511 rc2 = llu_md_close(sbi->ll_md_exp, inode);
519 * libsysio require us return 0
521 int llu_iop_close(struct inode *inode)
525 liblustre_wait_event(0);
527 rc = llu_file_release(inode);
529 CERROR("file close error %d\n", rc);
531 /* if open count == 0 && stale_flag is set, should we
532 * remove the inode immediately? */
533 liblustre_wait_idle();
537 _SYSIO_OFF_T llu_iop_pos(struct inode *ino, _SYSIO_OFF_T off)
541 liblustre_wait_event(0);
543 if (off < 0 || off > ll_file_maxbytes(ino))