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/super.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <sys/types.h>
50 #include <sys/queue.h>
52 # include <sys/statvfs.h>
54 # include <sys/statfs.h>
57 #include "llite_lib.h"
65 #define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH)
67 static int ll_permission(struct inode *inode, int mask)
69 struct intnl_stat *st = llu_i2stat(inode);
70 mode_t mode = st->st_mode;
72 if (current->fsuid == st->st_uid)
74 else if (in_group_p(st->st_gid))
77 if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
80 if ((mask & (MAY_READ|MAY_WRITE)) ||
81 (st->st_mode & S_IXUGO))
82 if (cfs_capable(CFS_CAP_DAC_OVERRIDE))
85 if (mask == MAY_READ ||
86 (S_ISDIR(st->st_mode) && !(mask & MAY_WRITE))) {
87 if (cfs_capable(CFS_CAP_DAC_READ_SEARCH))
94 static void llu_fsop_gone(struct filesys *fs)
96 struct llu_sb_info *sbi = (struct llu_sb_info *) fs->fs_private;
97 struct obd_device *obd = class_exp2obd(sbi->ll_md_exp);
101 cfs_list_del(&sbi->ll_conn_chain);
103 obd_disconnect(sbi->ll_dt_exp);
104 obd_disconnect(sbi->ll_md_exp);
106 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
107 class_manual_cleanup(obd);
109 OBD_FREE(sbi, sizeof(*sbi));
111 liblustre_wait_idle();
115 static struct inode_ops llu_inode_ops;
117 static ldlm_mode_t llu_take_md_lock(struct inode *inode, __u64 bits,
118 struct lustre_handle *lockh)
120 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
126 fid = &llu_i2info(inode)->lli_fid;
127 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
129 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
130 rc = md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
131 LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
135 void llu_update_inode(struct inode *inode, struct lustre_md *md)
137 struct llu_inode_info *lli = llu_i2info(inode);
138 struct mdt_body *body = md->body;
139 struct lov_stripe_md *lsm = md->lsm;
140 struct intnl_stat *st = llu_i2stat(inode);
142 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
144 if (body->valid & OBD_MD_FLMODE)
145 st->st_mode = (st->st_mode & S_IFMT)|(body->mode & ~S_IFMT);
146 if (body->valid & OBD_MD_FLTYPE)
147 st->st_mode = (st->st_mode & ~S_IFMT)|(body->mode & S_IFMT);
150 if (!lli->lli_has_smd) {
151 cl_file_inode_init(inode, md);
152 lli->lli_has_smd = true;
153 lli->lli_maxbytes = lsm->lsm_maxbytes;
154 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
155 lli->lli_maxbytes = MAX_LFS_FILESIZE;
158 obd_free_memmd(llu_i2obdexp(inode), &md->lsm);
161 if (body->valid & OBD_MD_FLATIME) {
162 if (body->atime > LTIME_S(st->st_atime))
163 LTIME_S(st->st_atime) = body->atime;
164 lli->lli_lvb.lvb_atime = body->atime;
166 if (body->valid & OBD_MD_FLMTIME) {
167 if (body->mtime > LTIME_S(st->st_mtime))
168 LTIME_S(st->st_mtime) = body->mtime;
169 lli->lli_lvb.lvb_mtime = body->mtime;
171 if (body->valid & OBD_MD_FLCTIME) {
172 if (body->ctime > LTIME_S(st->st_ctime))
173 LTIME_S(st->st_ctime) = body->ctime;
174 lli->lli_lvb.lvb_ctime = body->ctime;
176 if (S_ISREG(st->st_mode))
177 st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE);
179 st->st_blksize = 4096;
180 if (body->valid & OBD_MD_FLUID)
181 st->st_uid = body->uid;
182 if (body->valid & OBD_MD_FLGID)
183 st->st_gid = body->gid;
184 if (body->valid & OBD_MD_FLNLINK)
185 st->st_nlink = body->nlink;
186 if (body->valid & OBD_MD_FLRDEV)
187 st->st_rdev = body->rdev;
188 if (body->valid & OBD_MD_FLFLAGS)
189 lli->lli_st_flags = body->flags;
190 if (body->valid & OBD_MD_FLSIZE) {
191 if ((llu_i2sbi(inode)->ll_lco.lco_flags & OBD_CONNECT_SOM) &&
192 S_ISREG(st->st_mode) && lli->lli_has_smd) {
193 struct lustre_handle lockh;
196 /* As it is possible a blocking ast has been processed
197 * by this time, we need to check there is an UPDATE
198 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
200 mode = llu_take_md_lock(inode, MDS_INODELOCK_UPDATE,
203 st->st_size = body->size;
204 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
205 ldlm_lock_decref(&lockh, mode);
208 st->st_size = body->size;
211 if (body->valid & OBD_MD_FLBLOCKS)
212 st->st_blocks = body->blocks;
216 void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid)
218 struct llu_inode_info *lli = llu_i2info(dst);
219 struct intnl_stat *st = llu_i2stat(dst);
221 valid &= src->o_valid;
223 LASSERTF(!(valid & (OBD_MD_FLTYPE | OBD_MD_FLGENER | OBD_MD_FLFID |
224 OBD_MD_FLID | OBD_MD_FLGROUP)),
225 "object "DOSTID", valid %x\n", POSTID(&src->o_oi), valid);
227 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
228 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
231 LTIME_S(st->st_mtime), LTIME_S(st->st_ctime),
232 (long)src->o_mtime, (long)src->o_ctime);
234 if (valid & OBD_MD_FLATIME)
235 LTIME_S(st->st_atime) = src->o_atime;
236 if (valid & OBD_MD_FLMTIME)
237 LTIME_S(st->st_mtime) = src->o_mtime;
238 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
239 LTIME_S(st->st_ctime) = src->o_ctime;
240 if (valid & OBD_MD_FLSIZE)
241 st->st_size = src->o_size;
242 if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
243 st->st_blocks = src->o_blocks;
244 if (valid & OBD_MD_FLBLKSZ)
245 st->st_blksize = src->o_blksize;
246 if (valid & OBD_MD_FLTYPE)
247 st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
248 if (valid & OBD_MD_FLMODE)
249 st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
250 if (valid & OBD_MD_FLUID)
251 st->st_uid = src->o_uid;
252 if (valid & OBD_MD_FLGID)
253 st->st_gid = src->o_gid;
254 if (valid & OBD_MD_FLFLAGS)
255 lli->lli_st_flags = src->o_flags;
259 * Performs the getattr on the inode and updates its fields.
260 * If @sync != 0, perform the getattr under the server-side lock.
262 int llu_inode_getattr(struct inode *inode, struct obdo *obdo,
263 __u64 ioepoch, int sync)
265 struct ptlrpc_request_set *set;
266 struct lov_stripe_md *lsm = NULL;
267 struct obd_info oinfo = { { { 0 } } };
271 lsm = ccc_inode_lsm_get(inode);
276 oinfo.oi_oa->o_oi = lsm->lsm_oi;
277 oinfo.oi_oa->o_mode = S_IFREG;
278 oinfo.oi_oa->o_ioepoch = ioepoch;
279 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
280 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
281 OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
282 OBD_MD_FLCTIME | OBD_MD_FLGROUP |
283 OBD_MD_FLATIME | OBD_MD_FLEPOCH;
284 obdo_set_parent_fid(oinfo.oi_oa, &llu_i2info(inode)->lli_fid);
286 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
287 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
290 set = ptlrpc_prep_set();
292 CERROR ("ENOMEM allocing request set\n");
295 rc = obd_getattr_async(llu_i2obdexp(inode), &oinfo, set);
297 rc = ptlrpc_set_wait(set);
298 ptlrpc_set_destroy(set);
300 ccc_inode_lsm_put(inode, lsm);
304 oinfo.oi_oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
305 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
308 obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
309 CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu, "
310 "blksize %llu\n", POSTID(&oinfo.oi_oa->o_oi),
311 (long long unsigned)llu_i2stat(inode)->st_size,
312 (long long unsigned)llu_i2stat(inode)->st_blocks,
313 (long long unsigned)llu_i2stat(inode)->st_blksize);
317 static struct inode* llu_new_inode(struct filesys *fs,
321 struct llu_inode_info *lli;
322 struct intnl_stat st = {
325 #ifndef AUTOMOUNT_FILE_NAME
326 .st_mode = fid->f_type & S_IFMT,
328 .st_mode = fid->f_type /* all of the bits! */
331 /* FIXME: fix this later */
338 OBD_ALLOC(lli, sizeof(*lli));
342 /* initialize lli here */
343 lli->lli_sbi = llu_fs2sbi(fs);
344 lli->lli_has_smd = false;
345 lli->lli_symlink_name = NULL;
347 lli->lli_maxbytes = (__u64)(~0UL);
348 lli->lli_file_data = NULL;
350 lli->lli_sysio_fid.fid_data = &lli->lli_fid;
351 lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
354 /* file identifier is needed by functions like _sysio_i_find() */
355 inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
356 &st, 0, &llu_inode_ops, lli);
359 OBD_FREE(lli, sizeof(*lli));
364 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
366 struct lustre_handle lockh;
367 ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
374 fid = &llu_i2info(inode)->lli_fid;
375 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
377 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
378 if (md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
379 LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
385 static int llu_inode_revalidate(struct inode *inode)
387 struct llu_inode_info *lli = llu_i2info(inode);
388 struct intnl_stat *st = llu_i2stat(inode);
391 if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
393 struct ptlrpc_request *req = NULL;
394 struct llu_sb_info *sbi = llu_i2sbi(inode);
395 struct md_op_data op_data = { { 0 } };
396 unsigned long valid = OBD_MD_FLGETATTR;
399 /* Why don't we update all valid MDS fields here, if we're
400 * doing an RPC anyways? -phil */
401 if (S_ISREG(st->st_mode)) {
402 ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
403 valid |= OBD_MD_FLEASIZE;
406 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen,
408 op_data.op_valid = valid;
410 rc = md_getattr(sbi->ll_md_exp, &op_data, &req);
412 CERROR("failure %d inode %llu\n", rc,
413 (long long)st->st_ino);
416 rc = md_get_lustre_md(sbi->ll_md_exp, req,
417 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
419 /* XXX Too paranoid? */
420 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
421 !((md.body->valid & OBD_MD_FLNLINK) &&
422 (md.body->nlink == 0))) {
423 CERROR("Asked for %s eadata but got %s (%d)\n",
424 (valid & OBD_MD_FLEASIZE) ? "some" : "no",
425 (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
426 md.body->eadatasize);
429 ptlrpc_req_finished(req);
434 llu_update_inode(inode, &md);
436 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
437 ptlrpc_req_finished(req);
440 if (!lli->lli_has_smd) {
441 /* object not yet allocated, don't validate size */
442 st->st_atime = lli->lli_lvb.lvb_atime;
443 st->st_mtime = lli->lli_lvb.lvb_mtime;
444 st->st_ctime = lli->lli_lvb.lvb_ctime;
448 /* ll_glimpse_size will prefer locally cached writes if they extend
450 RETURN(cl_glimpse_size(inode));
453 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
455 *b = *llu_i2stat(ino);
458 static int llu_iop_getattr(struct pnode *pno,
460 struct intnl_stat *b)
465 liblustre_wait_event(0);
469 LASSERT(pno->p_base->pb_ino);
470 ino = pno->p_base->pb_ino;
472 LASSERT(!pno || pno->p_base->pb_ino == ino);
475 /* libsysio might call us directly without intent lock,
476 * we must re-fetch the attrs here
478 rc = llu_inode_revalidate(ino);
480 copy_stat_buf(ino, b);
481 LASSERT(!llu_i2info(ino)->lli_it);
484 liblustre_wait_event(0);
488 static int null_if_equal(struct ldlm_lock *lock, void *data)
490 if (data == lock->l_ast_data) {
491 lock->l_ast_data = NULL;
493 if (lock->l_req_mode != lock->l_granted_mode)
494 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
497 return LDLM_ITER_CONTINUE;
500 void llu_clear_inode(struct inode *inode)
502 struct llu_inode_info *lli = llu_i2info(inode);
503 struct llu_sb_info *sbi = llu_i2sbi(inode);
504 struct lov_stripe_md *lsm;
507 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
508 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
511 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
512 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
514 lsm = ccc_inode_lsm_get(inode);
516 obd_change_cbdata(sbi->ll_dt_exp, lsm, null_if_equal, inode);
517 ccc_inode_lsm_put(inode, lsm);
519 cl_inode_fini(inode);
520 lli->lli_has_smd = false;
522 if (lli->lli_symlink_name) {
523 OBD_FREE(lli->lli_symlink_name,
524 strlen(lli->lli_symlink_name) + 1);
525 lli->lli_symlink_name = NULL;
531 void llu_iop_gone(struct inode *inode)
533 struct llu_inode_info *lli = llu_i2info(inode);
536 liblustre_wait_event(0);
537 llu_clear_inode(inode);
539 OBD_FREE(lli, sizeof(*lli));
543 static int inode_setattr(struct inode * inode, struct iattr * attr)
545 unsigned int ia_valid = attr->ia_valid;
546 struct intnl_stat *st = llu_i2stat(inode);
550 * inode_setattr() is only ever invoked with ATTR_SIZE (by
551 * llu_setattr_raw()) when file has no bodies. Check this.
553 LASSERT(ergo(ia_valid & ATTR_SIZE, !llu_i2info(inode)->lli_has_smd));
555 if (ia_valid & ATTR_SIZE)
556 st->st_size = attr->ia_size;
557 if (ia_valid & ATTR_UID)
558 st->st_uid = attr->ia_uid;
559 if (ia_valid & ATTR_GID)
560 st->st_gid = attr->ia_gid;
561 if (ia_valid & ATTR_ATIME)
562 st->st_atime = attr->ia_atime;
563 if (ia_valid & ATTR_MTIME)
564 st->st_mtime = attr->ia_mtime;
565 if (ia_valid & ATTR_CTIME)
566 st->st_ctime = attr->ia_ctime;
567 if (ia_valid & ATTR_MODE) {
568 st->st_mode = attr->ia_mode;
569 if (!in_group_p(st->st_gid) &&
570 !cfs_capable(CFS_CAP_FSETID))
571 st->st_mode &= ~S_ISGID;
573 /* mark_inode_dirty(inode); */
577 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
578 struct md_open_data **mod)
581 struct llu_sb_info *sbi = llu_i2sbi(inode);
582 struct ptlrpc_request *request = NULL;
586 llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
587 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
591 ptlrpc_req_finished(request);
592 if (rc != -EPERM && rc != -EACCES)
593 CERROR("md_setattr fails: rc = %d\n", rc);
597 rc = md_get_lustre_md(sbi->ll_md_exp, request,
598 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
600 ptlrpc_req_finished(request);
604 /* We call inode_setattr to adjust timestamps.
605 * If there is at least some data in file, we cleared ATTR_SIZE
606 * above to avoid invoking vmtruncate, otherwise it is important
607 * to call vmtruncate in inode_setattr to update inode->i_size
609 inode_setattr(inode, &op_data->op_attr);
610 llu_update_inode(inode, &md);
611 ptlrpc_req_finished(request);
616 /* Close IO epoch and send Size-on-MDS attribute update. */
617 static int llu_setattr_done_writing(struct inode *inode,
618 struct md_op_data *op_data,
619 struct md_open_data *mod)
621 struct llu_inode_info *lli = llu_i2info(inode);
622 struct intnl_stat *st = llu_i2stat(inode);
626 LASSERT(op_data != NULL);
627 if (!S_ISREG(st->st_mode))
630 /* XXX: pass och here for the recovery purpose. */
631 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
632 op_data->op_ioepoch, PFID(&lli->lli_fid));
634 op_data->op_flags = MF_EPOCH_CLOSE;
635 llu_done_writing_attr(inode, op_data);
636 llu_pack_inode2opdata(inode, op_data, NULL);
638 rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
640 /* MDS has instructed us to obtain Size-on-MDS attribute
641 * from OSTs and send setattr to back to MDS. */
642 rc = llu_som_update(inode, op_data);
644 CERROR("inode %llu mdc truncate failed: rc = %d\n",
645 (unsigned long long)st->st_ino, rc);
650 /* If this inode has objects allocated to it (lsm != NULL), then the OST
651 * object(s) determine the file size and mtime. Otherwise, the MDS will
652 * keep these values until such a time that objects are allocated for it.
653 * We do the MDS operations first, as it is checking permissions for us.
654 * We don't to the MDS RPC if there is nothing that we want to store there,
655 * otherwise there is no harm in updating mtime/atime on the MDS if we are
656 * going to do an RPC anyways.
658 * If we are doing a truncate, we will send the mtime and ctime updates
659 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
660 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
663 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
665 int has_lsm = llu_i2info(inode)->lli_has_smd;
666 struct intnl_stat *st = llu_i2stat(inode);
667 int ia_valid = attr->ia_valid;
668 struct md_op_data op_data = { { 0 } };
669 struct md_open_data *mod = NULL;
673 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
675 if (ia_valid & ATTR_SIZE) {
676 if (attr->ia_size > ll_file_maxbytes(inode)) {
677 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
678 (long long)attr->ia_size,
679 ll_file_maxbytes(inode));
683 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
686 /* We mark all of the fields "set" so MDS/OST does not re-set them */
687 if (attr->ia_valid & ATTR_CTIME) {
688 attr->ia_ctime = CFS_CURRENT_TIME;
689 attr->ia_valid |= ATTR_CTIME_SET;
691 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
692 attr->ia_atime = CFS_CURRENT_TIME;
693 attr->ia_valid |= ATTR_ATIME_SET;
695 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
696 attr->ia_mtime = CFS_CURRENT_TIME;
697 attr->ia_valid |= ATTR_MTIME_SET;
700 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
701 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
702 ", now = "CFS_TIME_T"\n",
703 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
704 LTIME_S(CFS_CURRENT_TIME));
706 /* NB: ATTR_SIZE will only be set after this point if the size
707 * resides on the MDS, ie, this file has no objects. */
709 attr->ia_valid &= ~ATTR_SIZE;
711 /* If only OST attributes being set on objects, don't do MDS RPC.
712 * In that case, we need to check permissions and update the local
713 * inode ourselves so we can call obdo_from_inode() always. */
714 if (ia_valid & (has_lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
715 memcpy(&op_data.op_attr, attr, sizeof(*attr));
717 /* Open epoch for truncate. */
718 if (exp_connect_som(llu_i2mdexp(inode)) &&
719 (ia_valid & ATTR_SIZE))
720 op_data.op_flags = MF_EPOCH_OPEN;
721 rc = llu_md_setattr(inode, &op_data, &mod);
725 llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch);
726 if (!has_lsm || !S_ISREG(st->st_mode)) {
727 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
731 /* The OST doesn't check permissions, but the alternative is
732 * a gratuitous RPC to the MDS. We already rely on the client
733 * to do read/write/truncate permission checks, so is mtime OK?
735 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
736 /* from sys_utime() */
737 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
738 if (current->fsuid != st->st_uid &&
739 (rc = ll_permission(inode, MAY_WRITE)) != 0)
742 /* from inode_change_ok() */
743 if (current->fsuid != st->st_uid &&
744 !cfs_capable(CFS_CAP_FOWNER))
750 /* Won't invoke llu_vmtruncate(), as we already cleared
752 inode_setattr(inode, attr);
755 if (ia_valid & ATTR_SIZE)
756 attr->ia_valid |= ATTR_SIZE;
757 if (ia_valid & (ATTR_SIZE |
758 ATTR_ATIME | ATTR_ATIME_SET |
759 ATTR_MTIME | ATTR_MTIME_SET))
760 /* on truncate and utimes send attributes to osts, setting
761 * mtime/atime to past will be performed under PW 0:EOF extent
762 * lock (new_size:EOF for truncate)
763 * it may seem excessive to send mtime/atime updates to osts
764 * when not setting times to past, but it is necessary due to
765 * possible time de-synchronization */
766 rc = cl_setattr_ost(inode, attr, NULL);
769 if (op_data.op_ioepoch)
770 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
771 return rc ? rc : rc1;
774 /* here we simply act as a thin layer to glue it with
775 * llu_setattr_raw(), which is copy from kernel
777 static int llu_iop_setattr(struct pnode *pno,
780 struct intnl_stat *stbuf)
786 liblustre_wait_event(0);
788 LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
789 SETATTR_UID | SETATTR_GID |
790 SETATTR_LEN | SETATTR_MODE)));
791 memset(&iattr, 0, sizeof(iattr));
793 if (mask & SETATTR_MODE) {
794 iattr.ia_mode = stbuf->st_mode;
795 iattr.ia_valid |= ATTR_MODE;
797 if (mask & SETATTR_MTIME) {
798 iattr.ia_mtime = stbuf->st_mtime;
799 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
801 if (mask & SETATTR_ATIME) {
802 iattr.ia_atime = stbuf->st_atime;
803 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
805 if (mask & SETATTR_UID) {
806 iattr.ia_uid = stbuf->st_uid;
807 iattr.ia_valid |= ATTR_UID;
809 if (mask & SETATTR_GID) {
810 iattr.ia_gid = stbuf->st_gid;
811 iattr.ia_valid |= ATTR_GID;
813 if (mask & SETATTR_LEN) {
814 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
815 iattr.ia_valid |= ATTR_SIZE;
818 iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
819 iattr.ia_ctime = CFS_CURRENT_TIME;
821 rc = llu_setattr_raw(ino, &iattr);
822 liblustre_wait_idle();
826 #define EXT2_LINK_MAX 32000
828 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
830 struct inode *dir = pno->p_base->pb_parent->pb_ino;
831 struct qstr *qstr = &pno->p_base->pb_name;
832 const char *name = qstr->name;
834 struct ptlrpc_request *request = NULL;
835 struct llu_sb_info *sbi = llu_i2sbi(dir);
836 struct md_op_data op_data = {{ 0 }};
840 liblustre_wait_event(0);
841 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
844 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
847 err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
848 S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
849 cfs_curproc_cap_pack(), 0, &request);
850 ptlrpc_req_finished(request);
851 liblustre_wait_event(0);
855 static int llu_readlink_internal(struct inode *inode,
856 struct ptlrpc_request **request,
859 struct llu_inode_info *lli = llu_i2info(inode);
860 struct llu_sb_info *sbi = llu_i2sbi(inode);
861 struct mdt_body *body;
862 struct intnl_stat *st = llu_i2stat(inode);
863 struct md_op_data op_data = {{ 0 }};
864 int rc, symlen = st->st_size + 1;
870 if (lli->lli_symlink_name) {
871 *symname = lli->lli_symlink_name;
872 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
876 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen,
878 op_data.op_valid = OBD_MD_LINKNAME;
880 rc = md_getattr(sbi->ll_md_exp, &op_data, request);
882 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
886 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
887 LASSERT(body != NULL);
889 if ((body->valid & OBD_MD_LINKNAME) == 0) {
890 CERROR ("OBD_MD_LINKNAME not set on reply\n");
891 GOTO (failed, rc = -EPROTO);
894 LASSERT(symlen != 0);
895 if (body->eadatasize != symlen) {
896 CERROR("inode %llu: symlink length %d not expected %d\n",
897 (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
898 GOTO(failed, rc = -EPROTO);
901 *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
902 if (*symname == NULL ||
903 strnlen(*symname, symlen) != symlen - 1) {
904 /* not full/NULL terminated */
905 CERROR("inode %llu: symlink not NULL terminated string"
906 "of length %d\n", (long long)st->st_ino, symlen - 1);
907 GOTO(failed, rc = -EPROTO);
910 OBD_ALLOC(lli->lli_symlink_name, symlen);
911 /* do not return an error if we cannot cache the symlink locally */
912 if (lli->lli_symlink_name)
913 memcpy(lli->lli_symlink_name, *symname, symlen);
918 ptlrpc_req_finished (*request);
922 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
924 struct inode *inode = pno->p_base->pb_ino;
925 struct ptlrpc_request *request;
930 liblustre_wait_event(0);
931 rc = llu_readlink_internal(inode, &request, &symname);
936 strncpy(data, symname, bufsize);
937 rc = strlen(symname);
939 ptlrpc_req_finished(request);
941 liblustre_wait_event(0);
945 static int llu_iop_mknod_raw(struct pnode *pno,
949 struct ptlrpc_request *request = NULL;
950 struct inode *dir = pno->p_parent->p_base->pb_ino;
951 struct llu_sb_info *sbi = llu_i2sbi(dir);
952 struct md_op_data op_data = {{ 0 }};
956 liblustre_wait_event(0);
957 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
958 (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
959 (long long)llu_i2stat(dir)->st_ino);
961 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
964 switch (mode & S_IFMT) {
967 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
972 llu_prep_md_op_data(&op_data, dir, NULL,
973 pno->p_base->pb_name.name,
974 pno->p_base->pb_name.len, 0,
977 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
978 current->fsuid, current->fsgid,
979 cfs_curproc_cap_pack(), dev, &request);
980 ptlrpc_req_finished(request);
988 liblustre_wait_event(0);
992 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
994 struct inode *src = old->p_base->pb_ino;
995 struct inode *dir = new->p_parent->p_base->pb_ino;
996 const char *name = new->p_base->pb_name.name;
997 int namelen = new->p_base->pb_name.len;
998 struct ptlrpc_request *request = NULL;
999 struct md_op_data op_data = {{ 0 }};
1006 liblustre_wait_event(0);
1007 llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1009 rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1010 ptlrpc_req_finished(request);
1011 liblustre_wait_event(0);
1017 * libsysio will clear the inode immediately after return
1019 static int llu_iop_unlink_raw(struct pnode *pno)
1021 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1022 struct qstr *qstr = &pno->p_base->pb_name;
1023 const char *name = qstr->name;
1024 int len = qstr->len;
1025 struct inode *target = pno->p_base->pb_ino;
1026 struct ptlrpc_request *request = NULL;
1027 struct md_op_data op_data = { { 0 } };
1033 liblustre_wait_event(0);
1034 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1036 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1038 rc = llu_objects_destroy(request, dir);
1039 ptlrpc_req_finished(request);
1040 liblustre_wait_idle();
1045 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1047 struct inode *src = old->p_parent->p_base->pb_ino;
1048 struct inode *tgt = new->p_parent->p_base->pb_ino;
1049 const char *oldname = old->p_base->pb_name.name;
1050 int oldnamelen = old->p_base->pb_name.len;
1051 const char *newname = new->p_base->pb_name.name;
1052 int newnamelen = new->p_base->pb_name.len;
1053 struct ptlrpc_request *request = NULL;
1054 struct md_op_data op_data = { { 0 } };
1061 liblustre_wait_event(0);
1062 llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1064 rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1065 oldname, oldnamelen, newname, newnamelen,
1068 rc = llu_objects_destroy(request, src);
1071 ptlrpc_req_finished(request);
1072 liblustre_wait_idle();
1077 #ifdef _HAVE_STATVFS
1078 static int llu_statfs_internal(struct llu_sb_info *sbi,
1079 struct obd_statfs *osfs, __u64 max_age)
1081 struct obd_statfs obd_osfs;
1085 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, 0);
1087 CERROR("md_statfs fails: rc = %d\n", rc);
1091 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1092 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1094 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1095 &obd_statfs, max_age, 0);
1097 CERROR("obd_statfs fails: rc = %d\n", rc);
1101 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1102 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1105 osfs->os_blocks = obd_osfs.os_blocks;
1106 osfs->os_bfree = obd_osfs.os_bfree;
1107 osfs->os_bavail = obd_osfs.os_bavail;
1109 /* If we don't have as many objects free on the OST as inodes
1110 * on the MDS, we reduce the total number of inodes to
1111 * compensate, so that the "inodes in use" number is correct.
1113 if (obd_osfs.os_ffree < osfs->os_ffree) {
1114 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1116 osfs->os_ffree = obd_osfs.os_ffree;
1122 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1124 struct obd_statfs osfs;
1127 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1129 /* For now we will always get up-to-date statfs values, but in the
1130 * future we may allow some amount of caching on the client (e.g.
1131 * from QOS or lprocfs updates). */
1132 rc = llu_statfs_internal(sbi, &osfs,
1133 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS));
1137 statfs_unpack(sfs, &osfs);
1139 if (sizeof(sfs->f_blocks) == 4) {
1140 while (osfs.os_blocks > ~0UL) {
1143 osfs.os_blocks >>= 1;
1144 osfs.os_bfree >>= 1;
1145 osfs.os_bavail >>= 1;
1149 sfs->f_blocks = osfs.os_blocks;
1150 sfs->f_bfree = osfs.os_bfree;
1151 sfs->f_bavail = osfs.os_bavail;
1156 static int llu_iop_statvfs(struct pnode *pno,
1158 struct intnl_statvfs *buf)
1164 liblustre_wait_event(0);
1167 LASSERT(pno->p_base->pb_ino);
1168 rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1172 /* from native driver */
1173 buf->f_bsize = fs.f_bsize; /* file system block size */
1174 buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1175 buf->f_blocks = fs.f_blocks;
1176 buf->f_bfree = fs.f_bfree;
1177 buf->f_bavail = fs.f_bavail;
1178 buf->f_files = fs.f_files; /* Total number serial numbers */
1179 buf->f_ffree = fs.f_ffree; /* Number free serial numbers */
1180 buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1181 buf->f_fsid = fs.f_fsid.__val[1];
1182 buf->f_flag = 0; /* No equiv in statfs; maybe use type? */
1183 buf->f_namemax = fs.f_namelen;
1186 liblustre_wait_event(0);
1189 #endif /* _HAVE_STATVFS */
1191 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1193 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1194 struct qstr *qstr = &pno->p_base->pb_name;
1195 const char *name = qstr->name;
1196 int len = qstr->len;
1197 struct ptlrpc_request *request = NULL;
1198 struct intnl_stat *st = llu_i2stat(dir);
1199 struct md_op_data op_data = {{ 0 }};
1203 liblustre_wait_event(0);
1204 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1205 (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1207 if (st->st_nlink >= EXT2_LINK_MAX)
1210 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1213 err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1214 mode | S_IFDIR, current->fsuid, current->fsgid,
1215 cfs_curproc_cap_pack(), 0, &request);
1216 ptlrpc_req_finished(request);
1217 liblustre_wait_event(0);
1221 static int llu_iop_rmdir_raw(struct pnode *pno)
1223 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1224 struct qstr *qstr = &pno->p_base->pb_name;
1225 const char *name = qstr->name;
1226 int len = qstr->len;
1227 struct ptlrpc_request *request = NULL;
1228 struct md_op_data op_data = {{ 0 }};
1232 liblustre_wait_event(0);
1233 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1234 (long long)llu_i2stat(dir)->st_ino,
1235 llu_i2info(dir)->lli_st_generation, dir);
1237 llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1239 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1240 ptlrpc_req_finished(request);
1242 liblustre_wait_event(0);
1247 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1249 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1251 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1253 /* refer to ll_file_flock() for details */
1254 static int llu_file_flock(struct inode *ino,
1256 struct file_lock *file_lock)
1258 struct llu_inode_info *lli = llu_i2info(ino);
1259 struct ldlm_res_id res_id =
1260 { .name = {fid_seq(&lli->lli_fid),
1261 fid_oid(&lli->lli_fid),
1262 fid_ver(&lli->lli_fid),
1264 struct ldlm_enqueue_info einfo = {
1265 .ei_type = LDLM_FLOCK,
1267 .ei_cb_cp = ldlm_flock_completion_ast,
1268 .ei_cbdata = file_lock,
1270 struct intnl_stat *st = llu_i2stat(ino);
1271 struct lustre_handle lockh = {0};
1272 ldlm_policy_data_t flock;
1276 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1277 (unsigned long long)st->st_ino, file_lock);
1279 flock.l_flock.pid = file_lock->fl_pid;
1280 flock.l_flock.start = file_lock->fl_start;
1281 flock.l_flock.end = file_lock->fl_end;
1283 switch (file_lock->fl_type) {
1285 einfo.ei_mode = LCK_PR;
1288 einfo.ei_mode = LCK_NL;
1291 einfo.ei_mode = LCK_PW;
1294 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1301 #if F_SETLKW64 != F_SETLKW
1309 #if F_SETLK64 != F_SETLK
1313 flags = LDLM_FL_BLOCK_NOWAIT;
1317 #if F_GETLK64 != F_GETLK
1321 flags = LDLM_FL_TEST_LOCK;
1322 file_lock->fl_type = einfo.ei_mode;
1325 CERROR("unknown fcntl cmd: %d\n", cmd);
1329 CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#llx, mode=%u, "
1330 "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1331 flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1335 struct lmv_obd *lmv;
1336 struct obd_device *lmv_obd;
1337 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1338 lmv = &lmv_obd->u.lmv;
1340 if (lmv->desc.ld_tgt_count < 1)
1341 RETURN(rc = -ENODEV);
1343 if (lmv->tgts[0] != NULL && lmv->tgts[0]->ltd_exp != NULL)
1344 rc = ldlm_cli_enqueue(lmv->tgts[0]->ltd_exp, NULL,
1345 &einfo, &res_id, &flock, &flags,
1346 NULL, 0, LVB_T_NONE, &lockh, 0);
1353 static int assign_type(struct file_lock *fl, int type)
1366 static int flock_to_posix_lock(struct inode *ino,
1367 struct file_lock *fl,
1370 switch (l->l_whence) {
1371 /* XXX: only SEEK_SET is supported in lustre */
1379 fl->fl_end = l->l_len - 1;
1383 fl->fl_end = OFFSET_MAX;
1385 fl->fl_pid = getpid();
1386 fl->fl_flags = FL_POSIX;
1387 fl->fl_notify = NULL;
1388 fl->fl_insert = NULL;
1389 fl->fl_remove = NULL;
1390 /* XXX: these fields can't be filled with suitable values,
1391 but I think lustre doesn't use them.
1393 fl->fl_owner = NULL;
1396 return assign_type(fl, l->l_type);
1399 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1401 struct file_lock fl;
1405 if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1408 error = flock_to_posix_lock(ino, &fl, flock);
1412 error = llu_file_flock(ino, F_GETLK, &fl);
1416 flock->l_type = F_UNLCK;
1417 if (fl.fl_type != F_UNLCK) {
1418 flock->l_pid = fl.fl_pid;
1419 flock->l_start = fl.fl_start;
1420 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1421 fl.fl_end - fl.fl_start + 1;
1422 flock->l_whence = SEEK_SET;
1423 flock->l_type = fl.fl_type;
1430 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1432 struct file_lock fl;
1433 int flags = llu_i2info(ino)->lli_open_flags + 1;
1436 error = flock_to_posix_lock(ino, &fl, flock);
1439 if (cmd == F_SETLKW)
1440 fl.fl_flags |= FL_SLEEP;
1443 switch (flock->l_type) {
1445 if (!(flags & FMODE_READ))
1449 if (!(flags & FMODE_WRITE))
1459 error = llu_file_flock(ino, cmd, &fl);
1467 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1469 struct llu_inode_info *lli = llu_i2info(ino);
1471 struct flock *flock;
1474 liblustre_wait_event(0);
1477 *rtn = lli->lli_open_flags;
1480 flags = va_arg(ap, long);
1481 flags &= FCNTL_FLMASK;
1482 if (flags & FCNTL_FLMASK_INVALID) {
1483 LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1484 "the O_NONBLOCK or O_ASYNC flags. "
1485 "Please fix your application.\n");
1490 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1491 (lli->lli_open_flags & ~FCNTL_FLMASK);
1496 #if F_GETLK64 != F_GETLK
1500 flock = va_arg(ap, struct flock *);
1501 err = llu_fcntl_getlk(ino, flock);
1506 #if F_SETLKW64 != F_SETLKW
1512 #if F_SETLK64 != F_SETLK
1516 flock = va_arg(ap, struct flock *);
1517 err = llu_fcntl_setlk(ino, cmd, flock);
1521 CERROR("unsupported fcntl cmd %x\n", cmd);
1527 liblustre_wait_event(0);
1531 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1533 struct llu_inode_info *lli = llu_i2info(inode);
1534 struct ll_file_data *fd = lli->lli_file_data;
1536 struct ccc_grouplock grouplock;
1539 if (fd->fd_flags & LL_FILE_IGNORE_LOCK) {
1542 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1545 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1547 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1548 arg, (lli->lli_open_flags & O_NONBLOCK),
1554 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1555 fd->fd_grouplock = grouplock;
1560 int llu_put_grouplock(struct inode *inode, unsigned long arg)
1562 struct llu_inode_info *lli = llu_i2info(inode);
1563 struct ll_file_data *fd = lli->lli_file_data;
1564 struct ccc_grouplock grouplock;
1567 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1570 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1572 if (fd->fd_grouplock.cg_gid != arg)
1575 grouplock = fd->fd_grouplock;
1576 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1577 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1579 cl_put_grouplock(&grouplock);
1584 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1586 struct llu_sb_info *sbi = llu_i2sbi(ino);
1587 struct ptlrpc_request *request = NULL;
1588 struct md_op_data op_data = {{ 0 }};
1589 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1592 llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1595 LASSERT(sizeof(lum) == sizeof(*lump));
1596 LASSERT(sizeof(lum.lmm_objects[0]) ==
1597 sizeof(lump->lmm_objects[0]));
1598 if (copy_from_user(&lum, lump, sizeof(lum)))
1601 switch (lum.lmm_magic) {
1602 case LOV_USER_MAGIC_V1: {
1603 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1604 lustre_swab_lov_user_md_v1(&lum);
1607 case LOV_USER_MAGIC_V3: {
1608 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1609 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1613 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1614 " %#08x != %#08x nor %#08x\n",
1615 lum.lmm_magic, LOV_USER_MAGIC_V1,
1621 /* swabbing is done in lov_setstripe() on server side */
1622 rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1623 sizeof(lum), NULL, 0, &request, NULL);
1625 ptlrpc_req_finished(request);
1626 if (rc != -EPERM && rc != -EACCES)
1627 CERROR("md_setattr fails: rc = %d\n", rc);
1630 ptlrpc_req_finished(request);
1635 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1636 struct lov_user_md *lum, int lum_size)
1638 struct llu_sb_info *sbi = llu_i2sbi(ino);
1639 struct llu_inode_info *lli = llu_i2info(ino);
1640 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1641 struct ldlm_enqueue_info einfo = {
1642 .ei_type = LDLM_IBITS,
1644 .ei_cb_bl = llu_md_blocking_ast,
1645 .ei_cb_cp = ldlm_completion_ast,
1647 struct ptlrpc_request *req = NULL;
1648 struct lustre_md md;
1649 struct md_op_data data = {{ 0 }};
1650 struct lustre_handle lockh;
1654 if (lli->lli_has_smd) {
1655 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1656 PFID(&lli->lli_fid));
1660 llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1662 rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1663 &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1667 req = oit.d.lustre.it_data;
1668 rc = it_open_error(DISP_IT_EXECD, &oit);
1674 rc = it_open_error(DISP_OPEN_OPEN, &oit);
1680 rc = md_get_lustre_md(sbi->ll_md_exp, req,
1681 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1685 llu_update_inode(ino, &md);
1686 llu_local_open(lli, &oit);
1687 /* release intent */
1688 if (lustre_handle_is_used(&lockh))
1689 ldlm_lock_decref(&lockh, LCK_CR);
1690 ptlrpc_req_finished(req);
1692 rc = llu_file_release(ino);
1697 ptlrpc_req_finished(req);
1701 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1703 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1705 int flags = FMODE_WRITE;
1708 LASSERT(sizeof(lum) == sizeof(*lump));
1709 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1710 if (copy_from_user(&lum, lump, sizeof(lum)))
1713 rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1717 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1719 struct intnl_stat *st = llu_i2stat(ino);
1720 if (S_ISREG(st->st_mode))
1721 return llu_lov_file_setstripe(ino, arg);
1722 if (S_ISDIR(st->st_mode))
1723 return llu_lov_dir_setstripe(ino, arg);
1728 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1730 struct lov_stripe_md *lsm = NULL;
1733 lsm = ccc_inode_lsm_get(ino);
1735 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1737 ccc_inode_lsm_put(ino, lsm);
1741 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1747 liblustre_wait_event(0);
1750 case LL_IOC_GROUP_LOCK:
1751 arg = va_arg(ap, unsigned long);
1752 rc = llu_get_grouplock(ino, arg);
1754 case LL_IOC_GROUP_UNLOCK:
1755 arg = va_arg(ap, unsigned long);
1756 rc = llu_put_grouplock(ino, arg);
1758 case LL_IOC_LOV_SETSTRIPE:
1759 arg = va_arg(ap, unsigned long);
1760 rc = llu_lov_setstripe(ino, arg);
1762 case LL_IOC_LOV_GETSTRIPE:
1763 arg = va_arg(ap, unsigned long);
1764 rc = llu_lov_getstripe(ino, arg);
1767 CERROR("did not support ioctl cmd %lx\n", request);
1772 liblustre_wait_event(0);
1777 * we already do syncronous read/write
1779 static int llu_iop_sync(struct inode *inode)
1781 liblustre_wait_event(0);
1785 static int llu_iop_datasync(struct inode *inode)
1787 liblustre_wait_event(0);
1791 struct filesys_ops llu_filesys_ops =
1793 fsop_gone: llu_fsop_gone,
1796 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1798 struct inode *inode;
1800 struct file_identifier fileid = {&fid, sizeof(fid)};
1802 if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1803 (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1804 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1806 return ERR_PTR(-EPERM);
1809 /* try to find existing inode */
1810 fid = md->body->fid1;
1812 inode = _sysio_i_find(fs, &fileid);
1814 if (inode->i_zombie/* ||
1815 lli->lli_st_generation != md->body->generation*/) {
1819 llu_update_inode(inode, md);
1824 inode = llu_new_inode(fs, &fid);
1826 llu_update_inode(inode, md);
1832 llu_fsswop_mount(const char *source,
1834 const void *data __IS_UNUSED,
1835 struct pnode *tocover,
1836 struct mount **mntp)
1840 struct pnode_base *rootpb;
1841 struct obd_device *obd;
1842 struct llu_sb_info *sbi;
1843 struct obd_statfs osfs;
1844 static struct qstr noname = { NULL, 0, 0 };
1845 struct ptlrpc_request *request = NULL;
1846 struct lustre_md md;
1848 struct config_llog_instance cfg = {0, };
1849 struct lustre_profile *lprof;
1850 char *zconf_mgsnid, *zconf_profile;
1851 char *osc = NULL, *mdc = NULL;
1852 int async = 1, err = -EINVAL;
1853 struct obd_connect_data ocd = {0,};
1854 struct md_op_data op_data = {{0}};
1855 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1856 const int instlen = sizeof(cfg.cfg_instance) * 2 + 2;
1860 if (ll_parse_mount_target(source,
1863 CERROR("mal-formed target %s\n", source);
1866 if (!zconf_mgsnid || !zconf_profile) {
1867 printf("Liblustre: invalid target %s\n", source);
1870 /* allocate & initialize sbi */
1871 OBD_ALLOC(sbi, sizeof(*sbi));
1875 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
1876 ll_generate_random_uuid(uuid);
1877 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1879 /* generate a string unique to this super, let's try
1880 the address of the super itself.*/
1881 cfg.cfg_instance = sbi;
1883 /* retrive & parse config log */
1884 cfg.cfg_uuid = sbi->ll_sb_uuid;
1885 err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
1887 CERROR("Unable to process log: %s\n", zconf_profile);
1888 GOTO(out_free, err);
1891 lprof = class_get_profile(zconf_profile);
1892 if (lprof == NULL) {
1893 CERROR("No profile found: %s\n", zconf_profile);
1894 GOTO(out_free, err = -EINVAL);
1896 OBD_ALLOC(osc, strlen(lprof->lp_dt) + instlen + 2);
1897 sprintf(osc, "%s-%p", lprof->lp_dt, cfg.cfg_instance);
1899 OBD_ALLOC(mdc, strlen(lprof->lp_md) + instlen + 2);
1900 sprintf(mdc, "%s-%p", lprof->lp_md, cfg.cfg_instance);
1904 GOTO(out_free, err = -EINVAL);
1908 GOTO(out_free, err = -EINVAL);
1911 fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1917 obd = class_name2obd(mdc);
1919 CERROR("MDC %s: not setup or attached\n", mdc);
1920 GOTO(out_free, err = -EINVAL);
1922 obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC),
1923 KEY_ASYNC, sizeof(async), &async, NULL);
1925 ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
1926 OBD_CONNECT_FID | OBD_CONNECT_AT |
1927 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
1928 OBD_CONNECT_LVB_TYPE;
1930 #ifdef LIBLUSTRE_POSIX_ACL
1931 ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
1933 ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
1934 ocd.ocd_version = LUSTRE_VERSION_CODE;
1937 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1939 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
1940 GOTO(out_free, err);
1943 err = obd_statfs(NULL, sbi->ll_md_exp, &osfs, 100000000, 0);
1948 * FIXME fill fs stat data into sbi here!!! FIXME
1952 obd = class_name2obd(osc);
1954 CERROR("OSC %s: not setup or attached\n", osc);
1955 GOTO(out_md, err = -EINVAL);
1957 obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC),
1958 KEY_ASYNC, sizeof(async), &async, NULL);
1960 obd->obd_upcall.onu_owner = &sbi->ll_lco;
1961 obd->obd_upcall.onu_upcall = cl_ocd_update;
1963 ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
1964 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
1965 OBD_CONNECT_FID | OBD_CONNECT_AT |
1966 OBD_CONNECT_FULL20 | OBD_CONNECT_EINPROGRESS |
1967 OBD_CONNECT_LVB_TYPE;
1969 ocd.ocd_version = LUSTRE_VERSION_CODE;
1970 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1972 CERROR("cannot connect to %s: rc = %d\n", osc, err);
1975 sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
1976 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
1977 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
1979 fid_zero(&sbi->ll_root_fid);
1980 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
1982 CERROR("cannot mds_connect: rc = %d\n", err);
1983 GOTO(out_lock_cn_cb, err);
1985 if (!fid_is_sane(&sbi->ll_root_fid)) {
1986 CERROR("Invalid root fid during mount\n");
1987 GOTO(out_lock_cn_cb, err = -EINVAL);
1989 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
1991 op_data.op_fid1 = sbi->ll_root_fid;
1992 op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
1993 /* fetch attr of root inode */
1994 err = md_getattr(sbi->ll_md_exp, &op_data, &request);
1996 CERROR("md_getattr failed for root: rc = %d\n", err);
1997 GOTO(out_lock_cn_cb, err);
2000 err = md_get_lustre_md(sbi->ll_md_exp, request,
2001 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2003 CERROR("failed to understand root inode md: rc = %d\n",err);
2004 GOTO(out_request, err);
2007 LASSERT(fid_is_sane(&sbi->ll_root_fid));
2009 root = llu_iget(fs, &md);
2010 if (!root || IS_ERR(root)) {
2011 CERROR("fail to generate root inode\n");
2012 GOTO(out_request, err = -EBADF);
2016 * Generate base path-node for root.
2018 rootpb = _sysio_pb_new(&noname, NULL, root);
2024 err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2026 _sysio_pb_gone(rootpb);
2032 ptlrpc_req_finished(request);
2034 CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2039 _sysio_i_gone(root);
2041 ptlrpc_req_finished(request);
2043 obd_disconnect(sbi->ll_dt_exp);
2045 obd_disconnect(sbi->ll_md_exp);
2048 OBD_FREE(osc, strlen(lprof->lp_dt) + instlen + 2);
2050 OBD_FREE(mdc, strlen(lprof->lp_md) + instlen + 2);
2052 OBD_FREE(sbi, sizeof(*sbi));
2053 liblustre_wait_idle();
2057 struct fssw_ops llu_fssw_ops = {
2061 static struct inode_ops llu_inode_ops = {
2062 inop_lookup: llu_iop_lookup,
2063 inop_getattr: llu_iop_getattr,
2064 inop_setattr: llu_iop_setattr,
2065 inop_filldirentries: llu_iop_filldirentries,
2066 inop_mkdir: llu_iop_mkdir_raw,
2067 inop_rmdir: llu_iop_rmdir_raw,
2068 inop_symlink: llu_iop_symlink_raw,
2069 inop_readlink: llu_iop_readlink,
2070 inop_open: llu_iop_open,
2071 inop_close: llu_iop_close,
2072 inop_link: llu_iop_link_raw,
2073 inop_unlink: llu_iop_unlink_raw,
2074 inop_rename: llu_iop_rename_raw,
2075 inop_pos: llu_iop_pos,
2076 inop_read: llu_iop_read,
2077 inop_write: llu_iop_write,
2078 inop_iodone: llu_iop_iodone,
2079 inop_fcntl: llu_iop_fcntl,
2080 inop_sync: llu_iop_sync,
2081 inop_datasync: llu_iop_datasync,
2082 inop_ioctl: llu_iop_ioctl,
2083 inop_mknod: llu_iop_mknod_raw,
2084 #ifdef _HAVE_STATVFS
2085 inop_statvfs: llu_iop_statvfs,
2087 inop_gone: llu_iop_gone,