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 * 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>
52 #include "llite_lib.h"
54 /* Pack the required supplementary groups into the supplied groups array.
55 * If we don't need to use the groups from the target inode(s) then we
56 * instead pack one or more groups from the user's supplementary group
57 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
58 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
61 LASSERT(suppgids != NULL);
63 if (cfs_curproc_is_in_groups(i1->i_stbuf.st_gid))
64 suppgids[0] = i1->i_stbuf.st_gid;
69 if (cfs_curproc_is_in_groups(i2->i_stbuf.st_gid))
70 suppgids[1] = i2->i_stbuf.st_gid;
78 void llu_prep_md_op_data(struct md_op_data *op_data, struct inode *i1,
79 struct inode *i2, const char *name, int namelen,
82 LASSERT(i1 != NULL || i2 != NULL);
86 ll_i2gids(op_data->op_suppgids, i1, i2);
87 op_data->op_fid1 = *ll_inode2fid(i1);
89 ll_i2gids(op_data->op_suppgids, i2, i1);
90 op_data->op_fid1 = *ll_inode2fid(i2);
94 op_data->op_fid2 = *ll_inode2fid(i2);
96 fid_zero(&op_data->op_fid2);
98 op_data->op_opc = opc;
99 op_data->op_name = name;
100 op_data->op_mode = mode;
101 op_data->op_namelen = namelen;
102 op_data->op_mod_time = CFS_CURRENT_TIME;
103 op_data->op_data = NULL;
106 void llu_finish_md_op_data(struct md_op_data *op_data)
108 OBD_FREE_PTR(op_data);
111 void obdo_refresh_inode(struct inode *dst,
115 struct intnl_stat *st = llu_i2stat(dst);
116 valid &= src->o_valid;
118 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
119 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
121 src->o_valid, LTIME_S(st->st_mtime),
122 LTIME_S(st->st_ctime),
123 (long)src->o_mtime, (long)src->o_ctime);
125 if (valid & OBD_MD_FLATIME && src->o_atime > LTIME_S(st->st_atime))
126 LTIME_S(st->st_atime) = src->o_atime;
127 if (valid & OBD_MD_FLMTIME && src->o_mtime > LTIME_S(st->st_mtime))
128 LTIME_S(st->st_mtime) = src->o_mtime;
129 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
130 LTIME_S(st->st_ctime) = src->o_ctime;
131 if (valid & OBD_MD_FLSIZE && src->o_size > st->st_size)
132 st->st_size = src->o_size;
133 /* optimum IO size */
134 if (valid & OBD_MD_FLBLKSZ)
135 st->st_blksize = src->o_blksize;
136 /* allocation of space */
137 if (valid & OBD_MD_FLBLOCKS && src->o_blocks > st->st_blocks)
138 st->st_blocks = src->o_blocks;
142 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
143 * not believe attributes if a few ioepoch holders exist. Attributes for
144 * previous ioepoch if new one is opened are also skipped by MDS.
146 void llu_ioepoch_open(struct llu_inode_info *lli, __u64 ioepoch)
148 if (ioepoch && lli->lli_ioepoch != ioepoch) {
149 lli->lli_ioepoch = ioepoch;
150 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for truncate\n",
151 ioepoch, PFID(&lli->lli_fid));
155 int llu_local_open(struct llu_inode_info *lli, struct lookup_intent *it)
157 struct ptlrpc_request *req = it->d.lustre.it_data;
158 struct ll_file_data *fd;
159 struct mdt_body *body;
162 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
163 LASSERT(body != NULL);
165 /* already opened? */
166 if (lli->lli_open_count++)
169 LASSERT(!lli->lli_file_data);
171 OBD_ALLOC(fd, sizeof(*fd));
172 /* We can't handle this well without reorganizing ll_file_open and
173 * ll_md_close, so don't even try right now. */
176 memcpy(&fd->fd_mds_och.och_fh, &body->handle, sizeof(body->handle));
177 fd->fd_mds_och.och_magic = OBD_CLIENT_HANDLE_MAGIC;
178 fd->fd_mds_och.och_fid = lli->lli_fid;
179 lli->lli_file_data = fd;
180 llu_ioepoch_open(lli, body->ioepoch);
181 md_set_open_replay_data(lli->lli_sbi->ll_md_exp,
182 &fd->fd_mds_och, it->d.lustre.it_data);
187 int llu_iop_open(struct pnode *pnode, int flags, mode_t mode)
189 struct inode *inode = pnode->p_base->pb_ino;
190 struct llu_inode_info *lli = llu_i2info(inode);
191 struct intnl_stat *st = llu_i2stat(inode);
192 struct ll_file_data *fd;
193 struct ptlrpc_request *request;
194 struct lookup_intent *it;
195 struct lov_stripe_md *lsm;
199 liblustre_wait_event(0);
201 /* don't do anything for '/' */
202 if (llu_is_root_inode(inode))
205 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
206 LL_GET_INTENT(inode, it);
208 if (!it->d.lustre.it_disposition) {
212 rc = it_open_error(DISP_OPEN_OPEN, it);
214 GOTO(out_release, rc);
216 rc = llu_local_open(lli, it);
220 if (!S_ISREG(st->st_mode))
221 GOTO(out_release, rc = 0);
223 fd = lli->lli_file_data;
227 flags &= ~O_LOV_DELAY_CREATE;
228 /*XXX: open_flags are overwritten and the previous ones are lost */
229 lli->lli_open_flags = flags & ~(O_CREAT | O_EXCL | O_TRUNC);
232 request = it->d.lustre.it_data;
233 ptlrpc_req_finished(request);
235 it->it_op_release(it);
236 OBD_FREE(it, sizeof(*it));
238 /* libsysio hasn't done anything for O_TRUNC. here we
239 * simply simulate it as open(...); truncate(...); */
240 if (rc == 0 && (flags & O_TRUNC) && S_ISREG(st->st_mode)) {
243 memset(&attr, 0, sizeof(attr));
245 attr.ia_valid |= ATTR_SIZE | ATTR_RAW;
246 rc = llu_setattr_raw(inode, &attr);
248 CERROR("error %d truncate in open()\n", rc);
251 liblustre_wait_event(0);
255 int llu_objects_destroy(struct ptlrpc_request *req, struct inode *dir)
257 struct mdt_body *body;
258 struct lov_mds_md *eadata;
259 struct lov_stripe_md *lsm = NULL;
260 struct obd_trans_info oti = { 0 };
265 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
267 if (!(body->valid & OBD_MD_FLEASIZE))
270 if (body->eadatasize == 0) {
271 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
272 GOTO(out, rc = -EPROTO);
275 /* The MDS sent back the EA because we unlinked the last reference
276 * to this file. Use this EA to unlink the objects on the OST.
277 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
278 * check it is complete and sensible. */
279 eadata = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD,
282 LASSERT(eadata != NULL);
284 rc = obd_unpackmd(llu_i2obdexp(dir), &lsm, eadata,body->eadatasize);
286 CERROR("obd_unpackmd: %d\n", rc);
289 LASSERT(rc >= sizeof(*lsm));
293 GOTO(out_free_memmd, rc = -ENOMEM);
295 oa->o_id = lsm->lsm_object_id;
296 oa->o_seq = lsm->lsm_object_seq;
297 oa->o_mode = body->mode & S_IFMT;
298 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
299 obdo_from_inode(oa, NULL, &llu_i2info(dir)->lli_fid, 0);
301 if (body->valid & OBD_MD_FLCOOKIE) {
302 oa->o_valid |= OBD_MD_FLCOOKIE;
304 req_capsule_server_sized_get(&req->rq_pill,
306 sizeof(struct llog_cookie) *
307 lsm->lsm_stripe_count);
308 if (oti.oti_logcookies == NULL) {
309 oa->o_valid &= ~OBD_MD_FLCOOKIE;
310 body->valid &= ~OBD_MD_FLCOOKIE;
314 rc = obd_destroy(llu_i2obdexp(dir), oa, lsm, &oti, NULL, NULL);
317 CERROR("obd destroy objid 0x"LPX64" error %d\n",
318 lsm->lsm_object_id, rc);
320 obd_free_memmd(llu_i2obdexp(dir), &lsm);
325 /** Cliens updates SOM attributes on MDS: obd_getattr and md_setattr. */
326 int llu_som_update(struct inode *inode, struct md_op_data *op_data)
328 struct llu_inode_info *lli = llu_i2info(inode);
329 struct llu_sb_info *sbi = llu_i2sbi(inode);
330 struct obdo oa = { 0 };
335 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
336 LASSERT(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM);
338 old_flags = op_data->op_flags;
339 op_data->op_flags = MF_SOM_CHANGE;
341 /* If inode is already in another epoch, skip getattr from OSTs. */
342 if (lli->lli_ioepoch == op_data->op_ioepoch) {
343 rc = llu_inode_getattr(inode, &oa, op_data->op_ioepoch,
344 old_flags & MF_GETATTR_LOCK);
348 CDEBUG(D_INODE, "objid "LPX64" is destroyed\n",
349 lli->lli_smd->lsm_object_id);
351 CERROR("inode_getattr failed (%d): unable to "
352 "send a Size-on-MDS attribute update "
353 "for inode %llu/%lu\n", rc,
354 (long long)llu_i2stat(inode)->st_ino,
355 lli->lli_st_generation);
357 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
358 PFID(&lli->lli_fid));
361 /* Install attributes into op_data. */
362 md_from_obdo(op_data, &oa, oa.o_valid);
365 rc = llu_md_setattr(inode, op_data, NULL);
369 void llu_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
370 struct lustre_handle *fh)
372 struct llu_inode_info *lli = llu_i2info(inode);
373 struct intnl_stat *st = llu_i2stat(inode);
376 op_data->op_fid1 = lli->lli_fid;
377 op_data->op_attr.ia_atime = st->st_atime;
378 op_data->op_attr.ia_mtime = st->st_mtime;
379 op_data->op_attr.ia_ctime = st->st_ctime;
380 op_data->op_attr.ia_size = st->st_size;
381 op_data->op_attr_blocks = st->st_blocks;
382 op_data->op_attr.ia_attr_flags = lli->lli_st_flags;
383 op_data->op_ioepoch = lli->lli_ioepoch;
385 op_data->op_handle = *fh;
389 /** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
390 void llu_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
392 struct llu_inode_info *lli = llu_i2info(inode);
395 op_data->op_flags |= MF_SOM_CHANGE;
397 /* Pack Size-on-MDS attributes if we are in IO
398 * epoch and attributes are valid. */
399 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
400 if (!cl_local_size(inode))
401 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
402 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
407 static void llu_prepare_close(struct inode *inode, struct md_op_data *op_data,
408 struct ll_file_data *fd)
410 struct obd_client_handle *och = &fd->fd_mds_och;
412 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
413 ATTR_MTIME_SET | ATTR_CTIME_SET;
415 if (fd->fd_flags & FMODE_WRITE) {
416 struct llu_sb_info *sbi = llu_i2sbi(inode);
417 if (!(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM) ||
418 !S_ISREG(llu_i2stat(inode)->st_mode)) {
419 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
421 /* Inode cannot be dirty. Close the epoch. */
422 op_data->op_flags |= MF_EPOCH_CLOSE;
423 /* XXX: Send SOM attributes only if they are really
425 llu_done_writing_attr(inode, op_data);
428 llu_pack_inode2opdata(inode, op_data, &och->och_fh);
429 llu_prep_md_op_data(op_data, inode, NULL, NULL,
430 0, 0, LUSTRE_OPC_ANY);
433 int llu_md_close(struct obd_export *md_exp, struct inode *inode)
435 struct llu_inode_info *lli = llu_i2info(inode);
436 struct ll_file_data *fd = lli->lli_file_data;
437 struct ptlrpc_request *req = NULL;
438 struct obd_client_handle *och = &fd->fd_mds_och;
439 struct intnl_stat *st = llu_i2stat(inode);
440 struct md_op_data op_data = { { 0 } };
444 /* clear group lock, if present */
445 if (fd->fd_flags & LL_FILE_GROUP_LOCKED)
446 llu_put_grouplock(inode, fd->fd_grouplock.cg_gid);
448 llu_prepare_close(inode, &op_data, fd);
449 rc = md_close(md_exp, &op_data, och->och_mod, &req);
451 /* We are the last writer, so the MDS has instructed us to get
452 * the file size and any write cookies, then close again. */
453 LASSERT(lli->lli_open_flags & FMODE_WRITE);
454 rc = llu_som_update(inode, &op_data);
456 CERROR("inode %llu mdc Size-on-MDS update failed: "
457 "rc = %d\n", (long long)st->st_ino, rc);
461 CERROR("inode %llu close failed: rc %d\n",
462 (long long)st->st_ino, rc);
464 rc = llu_objects_destroy(req, inode);
466 CERROR("inode %llu ll_objects destroy: rc = %d\n",
467 (long long)st->st_ino, rc);
470 md_clear_open_replay_data(md_exp, och);
471 ptlrpc_req_finished(req);
472 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
473 lli->lli_file_data = NULL;
474 OBD_FREE(fd, sizeof(*fd));
479 int llu_file_release(struct inode *inode)
481 struct ll_file_data *fd;
482 struct llu_sb_info *sbi = llu_i2sbi(inode);
483 struct llu_inode_info *lli = llu_i2info(inode);
487 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu\n",
488 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation);
490 if (llu_is_root_inode(inode))
493 /* still opened by others? */
494 if (--lli->lli_open_count)
497 fd = lli->lli_file_data;
498 if (!fd) /* no process opened the file after an mcreate */
501 rc2 = llu_md_close(sbi->ll_md_exp, inode);
509 * libsysio require us return 0
511 int llu_iop_close(struct inode *inode)
515 liblustre_wait_event(0);
517 rc = llu_file_release(inode);
519 CERROR("file close error %d\n", rc);
521 /* if open count == 0 && stale_flag is set, should we
522 * remove the inode immediately? */
523 liblustre_wait_idle();
527 _SYSIO_OFF_T llu_iop_pos(struct inode *ino, _SYSIO_OFF_T off)
531 liblustre_wait_event(0);
533 if (off < 0 || off > ll_file_maxbytes(ino))