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/file.c
40 * Lustre Light file operations
43 #define DEBUG_SUBSYSTEM S_LLITE
49 #include <sys/types.h>
51 #include <sys/queue.h>
54 #include "llite_lib.h"
56 /* Pack the required supplementary groups into the supplied groups array.
57 * If we don't need to use the groups from the target inode(s) then we
58 * instead pack one or more groups from the user's supplementary group
59 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
60 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
63 LASSERT(suppgids != NULL);
65 if (cfs_curproc_is_in_groups(i1->i_stbuf.st_gid))
66 suppgids[0] = i1->i_stbuf.st_gid;
71 if (cfs_curproc_is_in_groups(i2->i_stbuf.st_gid))
72 suppgids[1] = i2->i_stbuf.st_gid;
80 void llu_prep_md_op_data(struct md_op_data *op_data, struct inode *i1,
81 struct inode *i2, const char *name, int namelen,
84 LASSERT(i1 != NULL || i2 != NULL);
88 ll_i2gids(op_data->op_suppgids, i1, i2);
89 op_data->op_fid1 = *ll_inode2fid(i1);
91 ll_i2gids(op_data->op_suppgids, i2, i1);
92 op_data->op_fid1 = *ll_inode2fid(i2);
96 op_data->op_fid2 = *ll_inode2fid(i2);
98 fid_zero(&op_data->op_fid2);
100 op_data->op_opc = opc;
101 op_data->op_name = name;
102 op_data->op_mode = mode;
103 op_data->op_namelen = namelen;
104 op_data->op_mod_time = CFS_CURRENT_TIME;
105 op_data->op_data = NULL;
108 void obdo_refresh_inode(struct inode *dst,
112 struct intnl_stat *st = llu_i2stat(dst);
113 valid &= src->o_valid;
115 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
116 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
118 src->o_valid, LTIME_S(st->st_mtime),
119 LTIME_S(st->st_ctime),
120 (long)src->o_mtime, (long)src->o_ctime);
122 if (valid & OBD_MD_FLATIME && src->o_atime > LTIME_S(st->st_atime))
123 LTIME_S(st->st_atime) = src->o_atime;
124 if (valid & OBD_MD_FLMTIME && src->o_mtime > LTIME_S(st->st_mtime))
125 LTIME_S(st->st_mtime) = src->o_mtime;
126 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
127 LTIME_S(st->st_ctime) = src->o_ctime;
128 if (valid & OBD_MD_FLSIZE && src->o_size > st->st_size)
129 st->st_size = src->o_size;
130 /* optimum IO size */
131 if (valid & OBD_MD_FLBLKSZ)
132 st->st_blksize = src->o_blksize;
133 /* allocation of space */
134 if (valid & OBD_MD_FLBLOCKS && src->o_blocks > st->st_blocks)
135 st->st_blocks = src->o_blocks;
139 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
140 * not believe attributes if a few ioepoch holders exist. Attributes for
141 * previous ioepoch if new one is opened are also skipped by MDS.
143 void llu_ioepoch_open(struct llu_inode_info *lli, __u64 ioepoch)
145 if (ioepoch && lli->lli_ioepoch != ioepoch) {
146 lli->lli_ioepoch = ioepoch;
147 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for truncate\n",
148 ioepoch, PFID(&lli->lli_fid));
152 int llu_local_open(struct llu_inode_info *lli, struct lookup_intent *it)
154 struct ptlrpc_request *req = it->d.lustre.it_data;
155 struct ll_file_data *fd;
156 struct mdt_body *body;
159 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
160 LASSERT(body != NULL);
162 /* already opened? */
163 if (lli->lli_open_count++)
166 LASSERT(!lli->lli_file_data);
168 OBD_ALLOC(fd, sizeof(*fd));
169 /* We can't handle this well without reorganizing ll_file_open and
170 * ll_md_close, so don't even try right now. */
173 memcpy(&fd->fd_mds_och.och_fh, &body->handle, sizeof(body->handle));
174 fd->fd_mds_och.och_magic = OBD_CLIENT_HANDLE_MAGIC;
175 fd->fd_mds_och.och_fid = lli->lli_fid;
176 lli->lli_file_data = fd;
177 llu_ioepoch_open(lli, body->ioepoch);
178 md_set_open_replay_data(lli->lli_sbi->ll_md_exp,
179 &fd->fd_mds_och, it->d.lustre.it_data);
184 int llu_iop_open(struct pnode *pnode, int flags, mode_t mode)
186 struct inode *inode = pnode->p_base->pb_ino;
187 struct llu_inode_info *lli = llu_i2info(inode);
188 struct intnl_stat *st = llu_i2stat(inode);
189 struct ptlrpc_request *request;
190 struct lookup_intent *it;
191 struct lov_stripe_md *lsm;
195 liblustre_wait_event(0);
197 /* don't do anything for '/' */
198 if (llu_is_root_inode(inode))
201 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
202 LL_GET_INTENT(inode, it);
204 if (!it->d.lustre.it_disposition) {
208 rc = it_open_error(DISP_OPEN_OPEN, it);
210 GOTO(out_release, rc);
212 rc = llu_local_open(lli, it);
216 if (!S_ISREG(st->st_mode))
217 GOTO(out_release, rc = 0);
221 flags &= ~O_LOV_DELAY_CREATE;
222 /*XXX: open_flags are overwritten and the previous ones are lost */
223 lli->lli_open_flags = flags & ~(O_CREAT | O_EXCL | O_TRUNC);
226 request = it->d.lustre.it_data;
227 ptlrpc_req_finished(request);
229 it->it_op_release(it);
230 OBD_FREE(it, sizeof(*it));
232 /* libsysio hasn't done anything for O_TRUNC. here we
233 * simply simulate it as open(...); truncate(...); */
234 if (rc == 0 && (flags & O_TRUNC) && S_ISREG(st->st_mode)) {
237 memset(&attr, 0, sizeof(attr));
239 attr.ia_valid |= ATTR_SIZE | ATTR_RAW;
240 rc = llu_setattr_raw(inode, &attr);
242 CERROR("error %d truncate in open()\n", rc);
245 liblustre_wait_event(0);
249 int llu_objects_destroy(struct ptlrpc_request *req, struct inode *dir)
251 struct mdt_body *body;
252 struct lov_mds_md *eadata;
253 struct lov_stripe_md *lsm = NULL;
254 struct obd_trans_info oti = { 0 };
259 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
261 if (!(body->valid & OBD_MD_FLEASIZE))
264 if (body->eadatasize == 0) {
265 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
266 GOTO(out, rc = -EPROTO);
269 /* The MDS sent back the EA because we unlinked the last reference
270 * to this file. Use this EA to unlink the objects on the OST.
271 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
272 * check it is complete and sensible. */
273 eadata = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD,
276 LASSERT(eadata != NULL);
278 rc = obd_unpackmd(llu_i2obdexp(dir), &lsm, eadata,body->eadatasize);
280 CERROR("obd_unpackmd: %d\n", rc);
283 LASSERT(rc >= sizeof(*lsm));
287 GOTO(out_free_memmd, rc = -ENOMEM);
289 oa->o_id = lsm->lsm_object_id;
290 oa->o_seq = lsm->lsm_object_seq;
291 oa->o_mode = body->mode & S_IFMT;
292 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
293 obdo_set_parent_fid(oa, &llu_i2info(dir)->lli_fid);
294 if (body->valid & OBD_MD_FLCOOKIE) {
295 oa->o_valid |= OBD_MD_FLCOOKIE;
297 req_capsule_server_sized_get(&req->rq_pill,
299 sizeof(struct llog_cookie) *
300 lsm->lsm_stripe_count);
301 if (oti.oti_logcookies == NULL) {
302 oa->o_valid &= ~OBD_MD_FLCOOKIE;
303 body->valid &= ~OBD_MD_FLCOOKIE;
307 rc = obd_destroy(llu_i2obdexp(dir), oa, lsm, &oti, NULL, NULL);
310 CERROR("obd destroy objid 0x"LPX64" error %d\n",
311 lsm->lsm_object_id, rc);
313 obd_free_memmd(llu_i2obdexp(dir), &lsm);
318 /** Cliens updates SOM attributes on MDS: obd_getattr and md_setattr. */
319 int llu_som_update(struct inode *inode, struct md_op_data *op_data)
321 struct llu_inode_info *lli = llu_i2info(inode);
322 struct llu_sb_info *sbi = llu_i2sbi(inode);
323 struct obdo oa = { 0 };
328 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
329 LASSERT(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM);
331 old_flags = op_data->op_flags;
332 op_data->op_flags = MF_SOM_CHANGE;
334 /* If inode is already in another epoch, skip getattr from OSTs. */
335 if (lli->lli_ioepoch == op_data->op_ioepoch) {
336 rc = llu_inode_getattr(inode, &oa, op_data->op_ioepoch,
337 old_flags & MF_GETATTR_LOCK);
341 CDEBUG(D_INODE, "objid "LPX64" is destroyed\n",
342 lli->lli_smd->lsm_object_id);
344 CERROR("inode_getattr failed (%d): unable to "
345 "send a Size-on-MDS attribute update "
346 "for inode %llu/%lu\n", rc,
347 (long long)llu_i2stat(inode)->st_ino,
348 lli->lli_st_generation);
350 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
351 PFID(&lli->lli_fid));
354 /* Install attributes into op_data. */
355 md_from_obdo(op_data, &oa, oa.o_valid);
358 rc = llu_md_setattr(inode, op_data, NULL);
362 void llu_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
363 struct lustre_handle *fh)
365 struct llu_inode_info *lli = llu_i2info(inode);
366 struct intnl_stat *st = llu_i2stat(inode);
369 op_data->op_fid1 = lli->lli_fid;
370 op_data->op_attr.ia_atime = st->st_atime;
371 op_data->op_attr.ia_mtime = st->st_mtime;
372 op_data->op_attr.ia_ctime = st->st_ctime;
373 op_data->op_attr.ia_size = st->st_size;
374 op_data->op_attr_blocks = st->st_blocks;
375 op_data->op_attr.ia_attr_flags = lli->lli_st_flags;
376 op_data->op_ioepoch = lli->lli_ioepoch;
378 op_data->op_handle = *fh;
382 /** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
383 void llu_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
385 struct llu_inode_info *lli = llu_i2info(inode);
388 op_data->op_flags |= MF_SOM_CHANGE;
390 /* Pack Size-on-MDS attributes if we are in IO
391 * epoch and attributes are valid. */
392 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
393 if (!cl_local_size(inode))
394 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
395 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
400 static void llu_prepare_close(struct inode *inode, struct md_op_data *op_data,
401 struct ll_file_data *fd)
403 struct obd_client_handle *och = &fd->fd_mds_och;
405 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
406 ATTR_MTIME_SET | ATTR_CTIME_SET;
408 if (fd->fd_flags & FMODE_WRITE) {
409 struct llu_sb_info *sbi = llu_i2sbi(inode);
410 if (!(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM) ||
411 !S_ISREG(llu_i2stat(inode)->st_mode)) {
412 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
414 /* Inode cannot be dirty. Close the epoch. */
415 op_data->op_flags |= MF_EPOCH_CLOSE;
416 /* XXX: Send SOM attributes only if they are really
418 llu_done_writing_attr(inode, op_data);
421 llu_pack_inode2opdata(inode, op_data, &och->och_fh);
422 llu_prep_md_op_data(op_data, inode, NULL, NULL,
423 0, 0, LUSTRE_OPC_ANY);
426 int llu_md_close(struct obd_export *md_exp, struct inode *inode)
428 struct llu_inode_info *lli = llu_i2info(inode);
429 struct ll_file_data *fd = lli->lli_file_data;
430 struct ptlrpc_request *req = NULL;
431 struct obd_client_handle *och = &fd->fd_mds_och;
432 struct intnl_stat *st = llu_i2stat(inode);
433 struct md_op_data op_data = { { 0 } };
437 /* clear group lock, if present */
438 if (fd->fd_flags & LL_FILE_GROUP_LOCKED)
439 llu_put_grouplock(inode, fd->fd_grouplock.cg_gid);
441 llu_prepare_close(inode, &op_data, fd);
442 rc = md_close(md_exp, &op_data, och->och_mod, &req);
444 /* We are the last writer, so the MDS has instructed us to get
445 * the file size and any write cookies, then close again. */
446 LASSERT(lli->lli_open_flags & FMODE_WRITE);
447 rc = llu_som_update(inode, &op_data);
449 CERROR("inode %llu mdc Size-on-MDS update failed: "
450 "rc = %d\n", (long long)st->st_ino, rc);
454 CERROR("inode %llu close failed: rc %d\n",
455 (long long)st->st_ino, rc);
457 rc = llu_objects_destroy(req, inode);
459 CERROR("inode %llu ll_objects destroy: rc = %d\n",
460 (long long)st->st_ino, rc);
463 md_clear_open_replay_data(md_exp, och);
464 ptlrpc_req_finished(req);
465 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
466 lli->lli_file_data = NULL;
467 OBD_FREE(fd, sizeof(*fd));
472 int llu_file_release(struct inode *inode)
474 struct ll_file_data *fd;
475 struct llu_sb_info *sbi = llu_i2sbi(inode);
476 struct llu_inode_info *lli = llu_i2info(inode);
480 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu\n",
481 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation);
483 if (llu_is_root_inode(inode))
486 /* still opened by others? */
487 if (--lli->lli_open_count)
490 fd = lli->lli_file_data;
491 if (!fd) /* no process opened the file after an mcreate */
494 rc2 = llu_md_close(sbi->ll_md_exp, inode);
502 * libsysio require us return 0
504 int llu_iop_close(struct inode *inode)
508 liblustre_wait_event(0);
510 rc = llu_file_release(inode);
512 CERROR("file close error %d\n", rc);
514 /* if open count == 0 && stale_flag is set, should we
515 * remove the inode immediately? */
516 liblustre_wait_idle();
520 _SYSIO_OFF_T llu_iop_pos(struct inode *ino, _SYSIO_OFF_T off)
524 liblustre_wait_event(0);
526 if (off < 0 || off > ll_file_maxbytes(ino))