4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
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>
48 #include <libcfs/libcfs.h>
49 #include <lustre/lustre_idl.h>
50 #include <liblustre.h>
52 #include <lustre_lib.h>
53 #include <lustre_lite.h>
54 #include <lustre_mdc.h>
55 #include <lustre_net.h>
56 #include <lustre_req_layout.h>
58 #include <obd_class.h>
59 #include <obd_support.h>
60 #include "llite_lib.h"
62 /* Pack the required supplementary groups into the supplied groups array.
63 * If we don't need to use the groups from the target inode(s) then we
64 * instead pack one or more groups from the user's supplementary group
65 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
66 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
69 LASSERT(suppgids != NULL);
71 if (in_group_p(i1->i_stbuf.st_gid))
72 suppgids[0] = i1->i_stbuf.st_gid;
77 if (in_group_p(i2->i_stbuf.st_gid))
78 suppgids[1] = i2->i_stbuf.st_gid;
86 void llu_prep_md_op_data(struct md_op_data *op_data, struct inode *i1,
87 struct inode *i2, const char *name, int namelen,
90 LASSERT(i1 != NULL || i2 != NULL);
94 ll_i2gids(op_data->op_suppgids, i1, i2);
95 op_data->op_fid1 = *ll_inode2fid(i1);
97 ll_i2gids(op_data->op_suppgids, i2, i1);
98 op_data->op_fid1 = *ll_inode2fid(i2);
102 op_data->op_fid2 = *ll_inode2fid(i2);
104 fid_zero(&op_data->op_fid2);
106 op_data->op_name = name;
107 op_data->op_mode = mode;
108 op_data->op_namelen = namelen;
109 op_data->op_mod_time = CFS_CURRENT_TIME;
110 op_data->op_data = NULL;
113 void obdo_refresh_inode(struct inode *dst,
117 struct intnl_stat *st = llu_i2stat(dst);
118 valid &= src->o_valid;
120 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
121 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
123 src->o_valid, LTIME_S(st->st_mtime),
124 LTIME_S(st->st_ctime),
125 (long)src->o_mtime, (long)src->o_ctime);
127 if (valid & OBD_MD_FLATIME && src->o_atime > LTIME_S(st->st_atime))
128 LTIME_S(st->st_atime) = src->o_atime;
129 if (valid & OBD_MD_FLMTIME && src->o_mtime > LTIME_S(st->st_mtime))
130 LTIME_S(st->st_mtime) = src->o_mtime;
131 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
132 LTIME_S(st->st_ctime) = src->o_ctime;
133 if (valid & OBD_MD_FLSIZE && src->o_size > st->st_size)
134 st->st_size = src->o_size;
135 /* optimum IO size */
136 if (valid & OBD_MD_FLBLKSZ)
137 st->st_blksize = src->o_blksize;
138 /* allocation of space */
139 if (valid & OBD_MD_FLBLOCKS && src->o_blocks > st->st_blocks)
140 st->st_blocks = src->o_blocks;
144 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
145 * not believe attributes if a few ioepoch holders exist. Attributes for
146 * previous ioepoch if new one is opened are also skipped by MDS.
148 void llu_ioepoch_open(struct llu_inode_info *lli, __u64 ioepoch)
150 if (ioepoch && lli->lli_ioepoch != ioepoch) {
151 lli->lli_ioepoch = ioepoch;
152 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for truncate\n",
153 ioepoch, PFID(&lli->lli_fid));
157 int llu_local_open(struct llu_inode_info *lli, struct lookup_intent *it)
159 struct ptlrpc_request *req = it->d.lustre.it_data;
160 struct ll_file_data *fd;
161 struct mdt_body *body;
164 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
165 LASSERT(body != NULL);
167 /* already opened? */
168 if (lli->lli_open_count++)
171 LASSERT(!lli->lli_file_data);
173 OBD_ALLOC(fd, sizeof(*fd));
174 /* We can't handle this well without reorganizing ll_file_open and
175 * ll_md_close, so don't even try right now. */
178 memcpy(&fd->fd_mds_och.och_fh, &body->handle, sizeof(body->handle));
179 fd->fd_mds_och.och_magic = OBD_CLIENT_HANDLE_MAGIC;
180 fd->fd_mds_och.och_fid = lli->lli_fid;
181 lli->lli_file_data = fd;
182 llu_ioepoch_open(lli, body->ioepoch);
183 md_set_open_replay_data(lli->lli_sbi->ll_md_exp, &fd->fd_mds_och, it);
188 int llu_iop_open(struct pnode *pnode, int flags, mode_t mode)
190 struct inode *inode = pnode->p_base->pb_ino;
191 struct llu_inode_info *lli = llu_i2info(inode);
192 struct intnl_stat *st = llu_i2stat(inode);
193 struct ptlrpc_request *request;
194 struct lookup_intent *it;
198 liblustre_wait_event(0);
200 /* don't do anything for '/' */
201 if (llu_is_root_inode(inode))
204 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
205 LL_GET_INTENT(inode, it);
207 if (!it->d.lustre.it_disposition) {
211 rc = it_open_error(DISP_OPEN_OPEN, it);
213 GOTO(out_release, rc);
215 rc = llu_local_open(lli, it);
219 if (!S_ISREG(st->st_mode))
220 GOTO(out_release, rc = 0);
222 if (lli->lli_has_smd && cl_is_lov_delay_create(flags)) {
223 /* a bit ugly, but better than changing the open() API */
224 unsigned int tmp_flags = flags;
226 cl_lov_delay_create_clear(&tmp_flags);
229 /*XXX: open_flags are overwritten and the previous ones are lost */
230 lli->lli_open_flags = flags & ~(O_CREAT | O_EXCL | O_TRUNC);
233 request = it->d.lustre.it_data;
234 ptlrpc_req_finished(request);
236 it->it_op_release(it);
237 OBD_FREE(it, sizeof(*it));
239 /* libsysio hasn't done anything for O_TRUNC. here we
240 * simply simulate it as open(...); truncate(...); */
241 if (rc == 0 && (flags & O_TRUNC) && S_ISREG(st->st_mode)) {
244 memset(&attr, 0, sizeof(attr));
246 attr.ia_valid |= ATTR_SIZE | ATTR_RAW;
247 rc = llu_setattr_raw(inode, &attr);
249 CERROR("error %d truncate in open()\n", rc);
252 liblustre_wait_event(0);
256 int llu_objects_destroy(struct ptlrpc_request *req, struct inode *dir)
258 struct mdt_body *body;
259 struct lov_mds_md *eadata;
260 struct lov_stripe_md *lsm = NULL;
261 struct obd_trans_info oti = { 0 };
266 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
268 if (!(body->valid & OBD_MD_FLEASIZE))
271 if (body->eadatasize == 0) {
272 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
273 GOTO(out, rc = -EPROTO);
276 /* The MDS sent back the EA because we unlinked the last reference
277 * to this file. Use this EA to unlink the objects on the OST.
278 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
279 * check it is complete and sensible. */
280 eadata = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD,
283 LASSERT(eadata != NULL);
285 rc = obd_unpackmd(llu_i2obdexp(dir), &lsm, eadata,body->eadatasize);
287 CERROR("obd_unpackmd: %d\n", rc);
290 LASSERT(rc >= sizeof(*lsm));
294 GOTO(out_free_memmd, rc = -ENOMEM);
296 oa->o_oi = lsm->lsm_oi;
297 oa->o_mode = body->mode & S_IFMT;
298 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
299 obdo_set_parent_fid(oa, &llu_i2info(dir)->lli_fid);
300 if (body->valid & OBD_MD_FLCOOKIE) {
301 oa->o_valid |= OBD_MD_FLCOOKIE;
303 req_capsule_server_sized_get(&req->rq_pill,
305 sizeof(struct llog_cookie) *
306 lsm->lsm_stripe_count);
307 if (oti.oti_logcookies == NULL) {
308 oa->o_valid &= ~OBD_MD_FLCOOKIE;
309 body->valid &= ~OBD_MD_FLCOOKIE;
313 rc = obd_destroy(NULL, llu_i2obdexp(dir), oa, lsm, &oti, NULL, NULL);
316 CERROR("obd destroy objid "DOSTID" error %d\n",
317 POSTID(&lsm->lsm_oi), rc);
319 obd_free_memmd(llu_i2obdexp(dir), &lsm);
324 /** Cliens updates SOM attributes on MDS: obd_getattr and md_setattr. */
325 int llu_som_update(struct inode *inode, struct md_op_data *op_data)
327 struct llu_inode_info *lli = llu_i2info(inode);
328 struct llu_sb_info *sbi = llu_i2sbi(inode);
329 struct obdo oa = { 0 };
334 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
335 LASSERT(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM);
337 old_flags = op_data->op_flags;
338 op_data->op_flags = MF_SOM_CHANGE;
340 /* If inode is already in another epoch, skip getattr from OSTs. */
341 if (lli->lli_ioepoch == op_data->op_ioepoch) {
342 rc = llu_inode_getattr(inode, &oa, op_data->op_ioepoch,
343 old_flags & MF_GETATTR_LOCK);
347 CERROR("inode_getattr failed (%d): unable to "
348 "send a Size-on-MDS attribute update "
349 "for inode %llu/%lu\n", rc,
350 (long long)llu_i2stat(inode)->st_ino,
351 lli->lli_st_generation);
353 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
354 PFID(&lli->lli_fid));
357 /* Install attributes into op_data. */
358 md_from_obdo(op_data, &oa, oa.o_valid);
361 rc = llu_md_setattr(inode, op_data, NULL);
365 void llu_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
366 struct lustre_handle *fh)
368 struct llu_inode_info *lli = llu_i2info(inode);
369 struct intnl_stat *st = llu_i2stat(inode);
372 op_data->op_fid1 = lli->lli_fid;
373 op_data->op_attr.ia_atime = st->st_atime;
374 op_data->op_attr.ia_mtime = st->st_mtime;
375 op_data->op_attr.ia_ctime = st->st_ctime;
376 op_data->op_attr.ia_size = st->st_size;
377 op_data->op_attr_blocks = st->st_blocks;
378 op_data->op_attr.ia_attr_flags = lli->lli_st_flags;
379 op_data->op_ioepoch = lli->lli_ioepoch;
381 op_data->op_handle = *fh;
385 /** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
386 void llu_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
388 struct llu_inode_info *lli = llu_i2info(inode);
391 op_data->op_flags |= MF_SOM_CHANGE;
393 /* Pack Size-on-MDS attributes if we are in IO
394 * epoch and attributes are valid. */
395 LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
396 if (!cl_local_size(inode))
397 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
398 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
403 static void llu_prepare_close(struct inode *inode, struct md_op_data *op_data,
404 struct ll_file_data *fd)
406 struct obd_client_handle *och = &fd->fd_mds_och;
408 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
409 ATTR_MTIME_SET | ATTR_CTIME_SET;
411 if (fd->fd_flags & FMODE_WRITE) {
412 struct llu_sb_info *sbi = llu_i2sbi(inode);
413 if (!(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM) ||
414 !S_ISREG(llu_i2stat(inode)->st_mode)) {
415 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
417 /* Inode cannot be dirty. Close the epoch. */
418 op_data->op_flags |= MF_EPOCH_CLOSE;
419 /* XXX: Send SOM attributes only if they are really
421 llu_done_writing_attr(inode, op_data);
424 llu_pack_inode2opdata(inode, op_data, &och->och_fh);
425 llu_prep_md_op_data(op_data, inode, NULL, NULL,
426 0, 0, LUSTRE_OPC_ANY);
429 int llu_md_close(struct obd_export *md_exp, struct inode *inode)
431 struct llu_inode_info *lli = llu_i2info(inode);
432 struct ll_file_data *fd = lli->lli_file_data;
433 struct ptlrpc_request *req = NULL;
434 struct obd_client_handle *och = &fd->fd_mds_och;
435 struct intnl_stat *st = llu_i2stat(inode);
436 struct md_op_data op_data = { { 0 } };
440 /* clear group lock, if present */
441 if (fd->fd_flags & LL_FILE_GROUP_LOCKED)
442 llu_put_grouplock(inode, fd->fd_grouplock.cg_gid);
444 llu_prepare_close(inode, &op_data, fd);
445 rc = md_close(md_exp, &op_data, och->och_mod, &req);
447 /* We are the last writer, so the MDS has instructed us to get
448 * the file size and any write cookies, then close again. */
449 LASSERT(lli->lli_open_flags & FMODE_WRITE);
450 rc = llu_som_update(inode, &op_data);
452 CERROR("inode %llu mdc Size-on-MDS update failed: "
453 "rc = %d\n", (long long)st->st_ino, rc);
457 CERROR("inode %llu close failed: rc %d\n",
458 (long long)st->st_ino, rc);
460 rc = llu_objects_destroy(req, inode);
462 CERROR("inode %llu ll_objects destroy: rc = %d\n",
463 (long long)st->st_ino, rc);
466 md_clear_open_replay_data(md_exp, och);
467 ptlrpc_req_finished(req);
468 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
469 lli->lli_file_data = NULL;
470 OBD_FREE(fd, sizeof(*fd));
475 int llu_file_release(struct inode *inode)
477 struct ll_file_data *fd;
478 struct llu_sb_info *sbi = llu_i2sbi(inode);
479 struct llu_inode_info *lli = llu_i2info(inode);
483 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu\n",
484 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation);
486 if (llu_is_root_inode(inode))
489 /* still opened by others? */
490 if (--lli->lli_open_count)
493 fd = lli->lli_file_data;
494 if (!fd) /* no process opened the file after an mcreate */
497 rc2 = llu_md_close(sbi->ll_md_exp, inode);
505 * libsysio require us return 0
507 int llu_iop_close(struct inode *inode)
511 liblustre_wait_event(0);
513 rc = llu_file_release(inode);
515 CERROR("file close error %d\n", rc);
517 /* if open count == 0 && stale_flag is set, should we
518 * remove the inode immediately? */
519 liblustre_wait_idle();
523 _SYSIO_OFF_T llu_iop_pos(struct inode *ino, _SYSIO_OFF_T off)
527 liblustre_wait_event(0);
529 if (off < 0 || off > ll_file_maxbytes(ino))