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;
139 /* mtime is always updated with ctime, but can be set in past.
140 As write and utime(2) may happen within 1 second, and utime's
141 mtime has a priority over write's one, leave mtime from mds
142 for the same ctimes. */
143 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime)) {
144 LTIME_S(st->st_ctime) = src->o_ctime;
145 if (valid & OBD_MD_FLMTIME)
146 LTIME_S(st->st_mtime) = src->o_mtime;
148 if (valid & OBD_MD_FLSIZE && src->o_size > st->st_size)
149 st->st_size = src->o_size;
150 /* optimum IO size */
151 if (valid & OBD_MD_FLBLKSZ)
152 st->st_blksize = src->o_blksize;
153 /* allocation of space */
154 if (valid & OBD_MD_FLBLOCKS && src->o_blocks > st->st_blocks)
155 st->st_blocks = src->o_blocks;
159 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
160 * not believe attributes if a few ioepoch holders exist. Attributes for
161 * previous ioepoch if new one is opened are also skipped by MDS.
163 void llu_ioepoch_open(struct llu_inode_info *lli, __u64 ioepoch)
165 if (ioepoch && lli->lli_ioepoch != ioepoch) {
166 lli->lli_ioepoch = ioepoch;
167 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for truncate\n",
168 ioepoch, PFID(&lli->lli_fid));
172 int llu_local_open(struct llu_inode_info *lli, struct lookup_intent *it)
174 struct ptlrpc_request *req = it->d.lustre.it_data;
175 struct ll_file_data *fd;
176 struct mdt_body *body;
179 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
180 LASSERT(body != NULL);
182 /* already opened? */
183 if (lli->lli_open_count++)
186 LASSERT(!lli->lli_file_data);
188 OBD_ALLOC(fd, sizeof(*fd));
189 /* We can't handle this well without reorganizing ll_file_open and
190 * ll_md_close, so don't even try right now. */
193 memcpy(&fd->fd_mds_och.och_fh, &body->handle, sizeof(body->handle));
194 fd->fd_mds_och.och_magic = OBD_CLIENT_HANDLE_MAGIC;
195 fd->fd_mds_och.och_fid = lli->lli_fid;
196 lli->lli_file_data = fd;
197 llu_ioepoch_open(lli, body->ioepoch);
198 md_set_open_replay_data(lli->lli_sbi->ll_md_exp,
199 &fd->fd_mds_och, it->d.lustre.it_data);
204 int llu_iop_open(struct pnode *pnode, int flags, mode_t mode)
206 struct inode *inode = pnode->p_base->pb_ino;
207 struct llu_inode_info *lli = llu_i2info(inode);
208 struct intnl_stat *st = llu_i2stat(inode);
209 struct ll_file_data *fd;
210 struct ptlrpc_request *request;
211 struct lookup_intent *it;
212 struct lov_stripe_md *lsm;
216 liblustre_wait_event(0);
218 /* don't do anything for '/' */
219 if (llu_is_root_inode(inode))
222 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
223 LL_GET_INTENT(inode, it);
225 if (!it->d.lustre.it_disposition) {
229 rc = it_open_error(DISP_OPEN_OPEN, it);
231 GOTO(out_release, rc);
233 rc = llu_local_open(lli, it);
237 if (!S_ISREG(st->st_mode))
238 GOTO(out_release, rc = 0);
240 fd = lli->lli_file_data;
244 flags &= ~O_LOV_DELAY_CREATE;
245 /*XXX: open_flags are overwritten and the previous ones are lost */
246 lli->lli_open_flags = flags & ~(O_CREAT | O_EXCL | O_TRUNC);
249 request = it->d.lustre.it_data;
250 ptlrpc_req_finished(request);
252 it->it_op_release(it);
253 OBD_FREE(it, sizeof(*it));
255 /* libsysio hasn't done anything for O_TRUNC. here we
256 * simply simulate it as open(...); truncate(...); */
257 if (rc == 0 && (flags & O_TRUNC) && S_ISREG(st->st_mode)) {
260 memset(&attr, 0, sizeof(attr));
262 attr.ia_valid |= ATTR_SIZE | ATTR_RAW;
263 rc = llu_setattr_raw(inode, &attr);
265 CERROR("error %d truncate in open()\n", rc);
268 liblustre_wait_event(0);
272 int llu_objects_destroy(struct ptlrpc_request *req, struct inode *dir)
274 struct mdt_body *body;
275 struct lov_mds_md *eadata;
276 struct lov_stripe_md *lsm = NULL;
277 struct obd_trans_info oti = { 0 };
282 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
284 if (!(body->valid & OBD_MD_FLEASIZE))
287 if (body->eadatasize == 0) {
288 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
289 GOTO(out, rc = -EPROTO);
292 /* The MDS sent back the EA because we unlinked the last reference
293 * to this file. Use this EA to unlink the objects on the OST.
294 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
295 * check it is complete and sensible. */
296 eadata = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD,
299 LASSERT(eadata != NULL);
301 rc = obd_unpackmd(llu_i2obdexp(dir), &lsm, eadata,body->eadatasize);
303 CERROR("obd_unpackmd: %d\n", rc);
306 LASSERT(rc >= sizeof(*lsm));
310 GOTO(out_free_memmd, rc = -ENOMEM);
312 oa->o_id = lsm->lsm_object_id;
313 oa->o_gr = lsm->lsm_object_gr;
314 oa->o_mode = body->mode & S_IFMT;
315 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
317 if (body->valid & OBD_MD_FLCOOKIE) {
318 oa->o_valid |= OBD_MD_FLCOOKIE;
320 req_capsule_server_sized_get(&req->rq_pill,
322 sizeof(struct llog_cookie) *
323 lsm->lsm_stripe_count);
324 if (oti.oti_logcookies == NULL) {
325 oa->o_valid &= ~OBD_MD_FLCOOKIE;
326 body->valid &= ~OBD_MD_FLCOOKIE;
330 rc = obd_destroy(llu_i2obdexp(dir), oa, lsm, &oti, NULL, NULL);
333 CERROR("obd destroy objid 0x"LPX64" error %d\n",
334 lsm->lsm_object_id, rc);
336 obd_free_memmd(llu_i2obdexp(dir), &lsm);
341 /** Cliens updates SOM attributes on MDS: obd_getattr and md_setattr. */
342 int llu_som_update(struct inode *inode, struct md_op_data *op_data)
344 struct llu_inode_info *lli = llu_i2info(inode);
345 struct llu_sb_info *sbi = llu_i2sbi(inode);
346 struct obdo oa = { 0 };
351 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
352 LASSERT(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM);
354 old_flags = op_data->op_flags;
355 op_data->op_flags = MF_SOM_CHANGE;
357 /* If inode is already in another epoch, skip getattr from OSTs. */
358 if (lli->lli_ioepoch == op_data->op_ioepoch) {
359 rc = llu_inode_getattr(inode, &oa, op_data->op_ioepoch,
360 old_flags & MF_GETATTR_LOCK);
364 CDEBUG(D_INODE, "objid "LPX64" is destroyed\n",
365 lli->lli_smd->lsm_object_id);
367 CERROR("inode_getattr failed (%d): unable to "
368 "send a Size-on-MDS attribute update "
369 "for inode %llu/%lu\n", rc,
370 (long long)llu_i2stat(inode)->st_ino,
371 lli->lli_st_generation);
373 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
374 PFID(&lli->lli_fid));
377 /* Install attributes into op_data. */
378 md_from_obdo(op_data, &oa, oa.o_valid);
381 rc = llu_md_setattr(inode, op_data, NULL);
385 void llu_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
386 struct lustre_handle *fh)
388 struct llu_inode_info *lli = llu_i2info(inode);
389 struct intnl_stat *st = llu_i2stat(inode);
392 op_data->op_fid1 = lli->lli_fid;
393 op_data->op_attr.ia_atime = st->st_atime;
394 op_data->op_attr.ia_mtime = st->st_mtime;
395 op_data->op_attr.ia_ctime = st->st_ctime;
396 op_data->op_attr.ia_size = st->st_size;
397 op_data->op_attr_blocks = st->st_blocks;
398 op_data->op_attr.ia_attr_flags = lli->lli_st_flags;
399 op_data->op_ioepoch = lli->lli_ioepoch;
401 op_data->op_handle = *fh;
405 /** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
406 void llu_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
408 struct llu_inode_info *lli = llu_i2info(inode);
411 op_data->op_flags |= MF_SOM_CHANGE;
413 /* Pack Size-on-MDS attributes if we are in IO
414 * epoch and attributes are valid. */
415 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
416 if (!cl_local_size(inode))
417 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
418 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
423 static void llu_prepare_close(struct inode *inode, struct md_op_data *op_data,
424 struct ll_file_data *fd)
426 struct obd_client_handle *och = &fd->fd_mds_och;
428 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
429 ATTR_MTIME_SET | ATTR_CTIME_SET;
431 if (fd->fd_flags & FMODE_WRITE) {
432 struct llu_sb_info *sbi = llu_i2sbi(inode);
433 if (!(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM) ||
434 !S_ISREG(llu_i2stat(inode)->st_mode)) {
435 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
437 /* Inode cannot be dirty. Close the epoch. */
438 op_data->op_flags |= MF_EPOCH_CLOSE;
439 /* XXX: Send SOM attributes only if they are really
441 llu_done_writing_attr(inode, op_data);
444 llu_pack_inode2opdata(inode, op_data, &och->och_fh);
445 llu_prep_md_op_data(op_data, inode, NULL, NULL,
446 0, 0, LUSTRE_OPC_ANY);
449 int llu_md_close(struct obd_export *md_exp, struct inode *inode)
451 struct llu_inode_info *lli = llu_i2info(inode);
452 struct ll_file_data *fd = lli->lli_file_data;
453 struct ptlrpc_request *req = NULL;
454 struct obd_client_handle *och = &fd->fd_mds_och;
455 struct intnl_stat *st = llu_i2stat(inode);
456 struct md_op_data op_data = { { 0 } };
460 /* clear group lock, if present */
461 if (fd->fd_flags & LL_FILE_GROUP_LOCKED)
462 llu_put_grouplock(inode, fd->fd_grouplock.cg_gid);
464 llu_prepare_close(inode, &op_data, fd);
465 rc = md_close(md_exp, &op_data, och->och_mod, &req);
467 /* We are the last writer, so the MDS has instructed us to get
468 * the file size and any write cookies, then close again. */
469 LASSERT(lli->lli_open_flags & FMODE_WRITE);
470 rc = llu_som_update(inode, &op_data);
472 CERROR("inode %llu mdc Size-on-MDS update failed: "
473 "rc = %d\n", (long long)st->st_ino, rc);
477 CERROR("inode %llu close failed: rc %d\n",
478 (long long)st->st_ino, rc);
480 rc = llu_objects_destroy(req, inode);
482 CERROR("inode %llu ll_objects destroy: rc = %d\n",
483 (long long)st->st_ino, rc);
486 md_clear_open_replay_data(md_exp, och);
487 ptlrpc_req_finished(req);
488 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
489 lli->lli_file_data = NULL;
490 OBD_FREE(fd, sizeof(*fd));
495 int llu_file_release(struct inode *inode)
497 struct ll_file_data *fd;
498 struct llu_sb_info *sbi = llu_i2sbi(inode);
499 struct llu_inode_info *lli = llu_i2info(inode);
503 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu\n",
504 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation);
506 if (llu_is_root_inode(inode))
509 /* still opened by others? */
510 if (--lli->lli_open_count)
513 fd = lli->lli_file_data;
514 if (!fd) /* no process opened the file after an mcreate */
517 rc2 = llu_md_close(sbi->ll_md_exp, inode);
525 * libsysio require us return 0
527 int llu_iop_close(struct inode *inode)
531 liblustre_wait_event(0);
533 rc = llu_file_release(inode);
535 CERROR("file close error %d\n", rc);
537 /* if open count == 0 && stale_flag is set, should we
538 * remove the inode immediately? */
539 liblustre_wait_idle();
543 _SYSIO_OFF_T llu_iop_pos(struct inode *ino, _SYSIO_OFF_T off)
547 liblustre_wait_event(0);
549 if (off < 0 || off > ll_file_maxbytes(ino))