1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * Copyright (c) 2002-2004 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
30 #include <sys/types.h>
32 #include <sys/fcntl.h>
33 #include <sys/queue.h>
36 # include <sys/statvfs.h>
38 # include <sys/statfs.h>
53 #include "llite_lib.h"
61 #define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH)
63 static int ll_permission(struct inode *inode, int mask)
65 struct llu_inode_info *lli = llu_i2info(inode);
66 mode_t mode = lli->lli_st_mode;
68 if (current->fsuid == lli->lli_st_uid)
70 else if (in_group_p(lli->lli_st_gid))
73 if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
76 if ((mask & (MAY_READ|MAY_WRITE)) ||
77 (lli->lli_st_mode & S_IXUGO))
78 if (capable(CAP_DAC_OVERRIDE))
81 if (mask == MAY_READ ||
82 (S_ISDIR(lli->lli_st_mode) && !(mask & MAY_WRITE))) {
83 if (capable(CAP_DAC_READ_SEARCH))
90 static void llu_fsop_gone(struct filesys *fs)
92 struct llu_sb_info *sbi = (struct llu_sb_info *)fs->fs_private;
93 struct obd_device *obd = class_exp2obd(sbi->ll_md_exp);
94 struct lustre_cfg_bufs bufs;
95 struct lustre_cfg *lcfg;
99 list_del(&sbi->ll_conn_chain);
100 obd_disconnect(sbi->ll_dt_exp, 0);
101 obd_disconnect(sbi->ll_md_exp, 0);
103 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
107 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
108 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
109 err = class_process_config(lcfg);
111 CERROR("cleanup failed: %s\n", obd->obd_name);
114 lcfg->lcfg_command = LCFG_DETACH;
115 err = class_process_config(lcfg);
116 lustre_cfg_free(lcfg);
118 CERROR("detach failed: %s\n", obd->obd_name);
122 obd_disconnect(sbi->ll_md_exp, 0);
123 OBD_FREE(sbi, sizeof(*sbi));
127 struct inode_ops llu_inode_ops;
129 void llu_update_inode(struct inode *inode, struct mds_body *body,
130 struct lov_stripe_md *lsm)
132 struct llu_inode_info *lli = llu_i2info(inode);
134 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
136 if (lli->lli_smd == NULL) {
138 lli->lli_maxbytes = lsm->lsm_maxbytes;
139 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
140 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
142 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
143 CERROR("lsm mismatch for inode %ld\n",
150 id_assign_fid(&lli->lli_id, &body->id1);
152 if ((body->valid & OBD_MD_FLID) || (body->valid & OBD_MD_FLGENER))
153 id_assign_stc(&lli->lli_id, &body->id1);
154 if (body->valid & OBD_MD_FLID)
155 lli->lli_st_ino = id_ino(&body->id1);
156 if (body->valid & OBD_MD_FLGENER)
157 lli->lli_st_generation = id_gen(&body->id1);
159 if (body->valid & OBD_MD_FLATIME)
160 LTIME_S(lli->lli_st_atime) = body->atime;
161 if (body->valid & OBD_MD_FLMTIME)
162 LTIME_S(lli->lli_st_mtime) = body->mtime;
163 if (body->valid & OBD_MD_FLCTIME)
164 LTIME_S(lli->lli_st_ctime) = body->ctime;
165 if (body->valid & OBD_MD_FLMODE)
166 lli->lli_st_mode = (lli->lli_st_mode & S_IFMT)|(body->mode & ~S_IFMT);
167 if (body->valid & OBD_MD_FLTYPE)
168 lli->lli_st_mode = (lli->lli_st_mode & ~S_IFMT)|(body->mode & S_IFMT);
169 if (body->valid & OBD_MD_FLUID)
170 lli->lli_st_uid = body->uid;
171 if (body->valid & OBD_MD_FLGID)
172 lli->lli_st_gid = body->gid;
173 if (body->valid & OBD_MD_FLFLAGS)
174 lli->lli_st_flags = body->flags;
175 if (body->valid & OBD_MD_FLNLINK)
176 lli->lli_st_nlink = body->nlink;
177 if (body->valid & OBD_MD_FLRDEV)
178 lli->lli_st_rdev = body->rdev;
179 if (body->valid & OBD_MD_FLSIZE)
180 lli->lli_st_size = body->size;
181 if (body->valid & OBD_MD_FLBLOCKS)
182 lli->lli_st_blocks = body->blocks;
185 void obdo_to_inode(struct inode *dst, struct obdo *src, obd_valid valid)
187 struct llu_inode_info *lli = llu_i2info(dst);
189 valid &= src->o_valid;
191 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
192 CDEBUG(D_INODE, "valid "LPX64", cur time %lu/%lu, new %lu/%lu\n",
194 LTIME_S(lli->lli_st_mtime), LTIME_S(lli->lli_st_ctime),
195 (long)src->o_mtime, (long)src->o_ctime);
197 if (valid & OBD_MD_FLATIME)
198 LTIME_S(lli->lli_st_atime) = src->o_atime;
199 if (valid & OBD_MD_FLMTIME)
200 LTIME_S(lli->lli_st_mtime) = src->o_mtime;
201 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(lli->lli_st_ctime))
202 LTIME_S(lli->lli_st_ctime) = src->o_ctime;
203 if (valid & OBD_MD_FLSIZE)
204 lli->lli_st_size = src->o_size;
205 if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
206 lli->lli_st_blocks = src->o_blocks;
207 if (valid & OBD_MD_FLBLKSZ)
208 lli->lli_st_blksize = src->o_blksize;
209 if (valid & OBD_MD_FLTYPE)
210 lli->lli_st_mode = (lli->lli_st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
211 if (valid & OBD_MD_FLMODE)
212 lli->lli_st_mode = (lli->lli_st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
213 if (valid & OBD_MD_FLUID)
214 lli->lli_st_uid = src->o_uid;
215 if (valid & OBD_MD_FLGID)
216 lli->lli_st_gid = src->o_gid;
217 if (valid & OBD_MD_FLFLAGS)
218 lli->lli_st_flags = src->o_flags;
219 if (valid & OBD_MD_FLGENER)
220 lli->lli_st_generation = src->o_generation;
223 #define S_IRWXUGO (S_IRWXU|S_IRWXG|S_IRWXO)
224 #define S_IALLUGO (S_ISUID|S_ISGID|S_ISVTX|S_IRWXUGO)
226 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_valid valid)
228 struct llu_inode_info *lli = llu_i2info(src);
229 obd_valid newvalid = 0;
231 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
232 CDEBUG(D_INODE, "valid "LPX64", new time %lu/%lu\n",
233 valid, LTIME_S(lli->lli_st_mtime),
234 LTIME_S(lli->lli_st_ctime));
236 if (valid & OBD_MD_FLATIME) {
237 dst->o_atime = LTIME_S(lli->lli_st_atime);
238 newvalid |= OBD_MD_FLATIME;
240 if (valid & OBD_MD_FLMTIME) {
241 dst->o_mtime = LTIME_S(lli->lli_st_mtime);
242 newvalid |= OBD_MD_FLMTIME;
244 if (valid & OBD_MD_FLCTIME) {
245 dst->o_ctime = LTIME_S(lli->lli_st_ctime);
246 newvalid |= OBD_MD_FLCTIME;
248 if (valid & OBD_MD_FLSIZE) {
249 dst->o_size = lli->lli_st_size;
250 newvalid |= OBD_MD_FLSIZE;
252 if (valid & OBD_MD_FLBLOCKS) { /* allocation of space (x512 bytes) */
253 dst->o_blocks = lli->lli_st_blocks;
254 newvalid |= OBD_MD_FLBLOCKS;
256 if (valid & OBD_MD_FLBLKSZ) { /* optimal block size */
257 dst->o_blksize = lli->lli_st_blksize;
258 newvalid |= OBD_MD_FLBLKSZ;
260 if (valid & OBD_MD_FLTYPE) {
261 dst->o_mode = (dst->o_mode & S_IALLUGO)|(lli->lli_st_mode & S_IFMT);
262 newvalid |= OBD_MD_FLTYPE;
264 if (valid & OBD_MD_FLMODE) {
265 dst->o_mode = (dst->o_mode & S_IFMT)|(lli->lli_st_mode & S_IALLUGO);
266 newvalid |= OBD_MD_FLMODE;
268 if (valid & OBD_MD_FLUID) {
269 dst->o_uid = lli->lli_st_uid;
270 newvalid |= OBD_MD_FLUID;
272 if (valid & OBD_MD_FLGID) {
273 dst->o_gid = lli->lli_st_gid;
274 newvalid |= OBD_MD_FLGID;
276 if (valid & OBD_MD_FLFLAGS) {
277 dst->o_flags = lli->lli_st_flags;
278 newvalid |= OBD_MD_FLFLAGS;
280 if (valid & OBD_MD_FLGENER) {
281 dst->o_generation = lli->lli_st_generation;
282 newvalid |= OBD_MD_FLGENER;
285 dst->o_valid |= newvalid;
289 * really does the getattr on the inode and updates its fields
291 int llu_inode_getattr(struct inode *inode, struct lov_stripe_md *lsm)
293 struct llu_inode_info *lli = llu_i2info(inode);
294 struct obd_export *exp = llu_i2dtexp(inode);
295 struct ptlrpc_request_set *set;
297 obd_valid refresh_valid;
304 memset(&oa, 0, sizeof oa);
305 oa.o_id = lsm->lsm_object_id;
307 oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE |
308 OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
311 set = ptlrpc_prep_set();
313 CERROR ("ENOMEM allocing request set\n");
316 rc = obd_getattr_async(exp, &oa, lsm, set);
318 rc = ptlrpc_set_wait(set);
319 ptlrpc_set_destroy(set);
324 refresh_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
325 OBD_MD_FLCTIME | OBD_MD_FLSIZE;
327 /* We set this flag in commit write as we extend the file size. When
328 * the bit is set and the lock is canceled that covers the file size,
329 * we clear the bit. This is enough to protect the window where our
330 * local size extension is needed for writeback. However, it relies on
331 * behaviour that won't be true in the near future. This assumes that
332 * all getattr callers get extent locks, which they currnetly do. It
333 * also assumes that we only send discarding asts for {0,eof} truncates
334 * as is currently the case. This will have to be replaced by the
335 * proper eoc communication between clients and the ost, which is on
337 if (test_bit(LLI_F_PREFER_EXTENDED_SIZE, &lli->lli_flags)) {
338 if (oa.o_size < lli->lli_st_size)
339 refresh_valid &= ~OBD_MD_FLSIZE;
341 clear_bit(LLI_F_PREFER_EXTENDED_SIZE, &lli->lli_flags);
344 obdo_refresh_inode(inode, &oa, refresh_valid);
349 static struct inode *llu_new_inode(struct filesys *fs,
350 struct lustre_id *id)
353 struct intnl_stat stat;
354 struct llu_inode_info *lli;
356 OBD_ALLOC(lli, sizeof(*lli));
360 /* initialize lli here */
361 lli->lli_sbi = llu_fs2sbi(fs);
363 lli->lli_symlink_name = NULL;
365 lli->lli_maxbytes = (__u64)(~0UL);
366 lli->lli_file_data = NULL;
368 lli->lli_sysio_fid.fid_data = &lli->lli_id;
369 lli->lli_sysio_fid.fid_len = sizeof(lli->lli_id);
371 memcpy(&lli->lli_id, id, sizeof(*id));
373 #warning "fill @stat by desired attributes of new inode before using_sysio_i_new()"
374 memset(&stat, 0, sizeof(stat));
375 stat.st_ino = id_ino(id);
377 /* file identifier is needed by functions like _sysio_i_find() */
378 inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
379 &stat, 0, &llu_inode_ops, lli);
382 OBD_FREE(lli, sizeof(*lli));
387 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
389 struct llu_sb_info *sbi = llu_i2sbi(inode);
390 struct llu_inode_info *lli = llu_i2info(inode);
391 struct lustre_handle lockh;
392 struct ldlm_res_id res_id = { .name = {0} };
393 struct obd_device *obddev;
394 ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
400 obddev = sbi->ll_md_exp->exp_obd;
401 res_id.name[0] = id_fid(&lli->lli_id);
402 res_id.name[1] = id_group(&lli->lli_id);
404 CDEBUG(D_INFO, "trying to match res "LPU64"\n", res_id.name[0]);
406 /* FIXME use LDLM_FL_TEST_LOCK instead */
407 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
408 if (ldlm_lock_match(obddev->obd_namespace, flags, &res_id, LDLM_IBITS,
409 &policy, LCK_PR, &lockh)) {
410 ldlm_lock_decref(&lockh, LCK_PR);
414 if (ldlm_lock_match(obddev->obd_namespace, flags, &res_id, LDLM_IBITS,
415 &policy, LCK_PW, &lockh)) {
416 ldlm_lock_decref(&lockh, LCK_PW);
422 static int llu_inode_revalidate(struct inode *inode)
424 struct llu_inode_info *lli = llu_i2info(inode);
425 struct lov_stripe_md *lsm = NULL;
429 CERROR("REPORT THIS LINE TO PETER\n");
433 if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
435 struct ptlrpc_request *req = NULL;
436 struct llu_sb_info *sbi = llu_i2sbi(inode);
441 /* Why don't we update all valid MDS fields here, if we're doing
442 * an RPC anyways? -phil */
443 if (S_ISREG(lli->lli_st_mode)) {
444 ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
445 valid |= OBD_MD_FLEASIZE;
447 ll_inode2id(&id, inode);
449 /* XXX: capa is NULL here, is it correct? */
450 rc = mdc_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
451 0, ealen, NULL, &req);
453 CERROR("failure %d inode %lu\n", rc, lli->lli_st_ino);
456 rc = mdc_req2lustre_md(sbi->ll_md_exp, req, 0,
457 sbi->ll_dt_exp, &md);
459 /* XXX Too paranoid? */
460 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
461 !((md.body->valid & OBD_MD_FLNLINK) &&
462 (md.body->nlink == 0))) {
463 CERROR("Asked for %s eadata but got %s (%d)\n",
464 (valid & OBD_MD_FLEASIZE) ? "some" : "no",
465 (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
466 md.body->eadatasize);
469 ptlrpc_req_finished(req);
474 llu_update_inode(inode, md.body, md.lsm);
475 if (md.lsm != NULL && llu_i2info(inode)->lli_smd != md.lsm)
476 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
478 if (md.body->valid & OBD_MD_FLSIZE)
479 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
480 &llu_i2info(inode)->lli_flags);
481 ptlrpc_req_finished(req);
484 lsm = llu_i2info(inode)->lli_smd;
485 if (!lsm) /* object not yet allocated, don't validate size */
488 /* ll_glimpse_size will prefer locally cached writes if they extend
490 RETURN(llu_glimpse_size(inode));
493 static void copy_stat_buf_lli(struct llu_inode_info *lli,
494 struct intnl_stat *b)
496 b->st_dev = lli->lli_st_dev;
497 b->st_ino = lli->lli_st_ino;
498 b->st_mode = lli->lli_st_mode;
499 b->st_nlink = lli->lli_st_nlink;
500 b->st_uid = lli->lli_st_uid;
501 b->st_gid = lli->lli_st_gid;
502 b->st_rdev = lli->lli_st_rdev;
503 b->st_size = lli->lli_st_size;
504 b->st_blksize = lli->lli_st_blksize;
505 b->st_blocks = lli->lli_st_blocks;
506 b->st_atime = lli->lli_st_atime;
507 b->st_mtime = lli->lli_st_mtime;
508 b->st_ctime = lli->lli_st_ctime;
511 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
513 struct llu_inode_info *lli = llu_i2info(ino);
514 copy_stat_buf_lli(lli, b);
517 static int llu_iop_getattr(struct pnode *pno,
519 struct intnl_stat *b)
524 liblustre_wait_event(0);
528 LASSERT(pno->p_base->pb_ino);
529 ino = pno->p_base->pb_ino;
531 LASSERT(!pno || pno->p_base->pb_ino == ino);
534 /* libsysio might call us directly without intent lock,
535 * we must re-fetch the attrs here
537 rc = llu_inode_revalidate(ino);
539 copy_stat_buf(ino, b);
540 LASSERT(!llu_i2info(ino)->lli_it);
546 static int null_if_equal(struct ldlm_lock *lock, void *data)
548 if (data == lock->l_ast_data) {
549 lock->l_ast_data = NULL;
551 if (lock->l_req_mode != lock->l_granted_mode)
552 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
555 return LDLM_ITER_CONTINUE;
558 void llu_clear_inode(struct inode *inode)
561 struct llu_inode_info *lli = llu_i2info(inode);
562 struct llu_sb_info *sbi = llu_i2sbi(inode);
565 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%lu(%p)\n", lli->lli_st_ino,
566 lli->lli_st_generation, inode);
568 ll_inode2id(&id, inode);
569 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(lli->lli_flags));
570 mdc_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
573 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
574 null_if_equal, inode);
577 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
581 if (lli->lli_symlink_name) {
582 OBD_FREE(lli->lli_symlink_name,
583 strlen(lli->lli_symlink_name) + 1);
584 lli->lli_symlink_name = NULL;
590 void llu_iop_gone(struct inode *inode)
592 struct llu_inode_info *lli = llu_i2info(inode);
595 liblustre_wait_event(0);
596 llu_clear_inode(inode);
598 OBD_FREE(lli, sizeof(*lli));
602 static int inode_setattr(struct inode * inode, struct iattr * attr)
604 unsigned int ia_valid = attr->ia_valid;
605 struct llu_inode_info *lli = llu_i2info(inode);
608 if (ia_valid & ATTR_SIZE) {
609 error = llu_vmtruncate(inode, attr->ia_size);
614 if (ia_valid & ATTR_UID)
615 lli->lli_st_uid = attr->ia_uid;
616 if (ia_valid & ATTR_GID)
617 lli->lli_st_gid = attr->ia_gid;
618 if (ia_valid & ATTR_ATIME)
619 lli->lli_st_atime = attr->ia_atime;
620 if (ia_valid & ATTR_MTIME)
621 lli->lli_st_mtime = attr->ia_mtime;
622 if (ia_valid & ATTR_CTIME)
623 lli->lli_st_ctime = attr->ia_ctime;
624 if (ia_valid & ATTR_MODE) {
625 lli->lli_st_mode = attr->ia_mode;
626 if (!in_group_p(lli->lli_st_gid) && !capable(CAP_FSETID))
627 lli->lli_st_mode &= ~S_ISGID;
629 /* mark_inode_dirty(inode); */
634 /* If this inode has objects allocated to it (lsm != NULL), then the OST
635 * object(s) determine the file size and mtime. Otherwise, the MDS will
636 * keep these values until such a time that objects are allocated for it.
637 * We do the MDS operations first, as it is checking permissions for us.
638 * We don't to the MDS RPC if there is nothing that we want to store there,
639 * otherwise there is no harm in updating mtime/atime on the MDS if we are
640 * going to do an RPC anyways.
642 * If we are doing a truncate, we will send the mtime and ctime updates
643 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
644 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
647 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
649 struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd;
650 struct llu_sb_info *sbi = llu_i2sbi(inode);
651 struct llu_inode_info *lli = llu_i2info(inode);
652 struct ptlrpc_request *request = NULL;
653 struct mdc_op_data op_data;
654 int ia_valid = attr->ia_valid;
658 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", lli->lli_st_ino);
660 if (ia_valid & ATTR_SIZE) {
661 if (attr->ia_size > ll_file_maxbytes(inode)) {
662 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
663 attr->ia_size, ll_file_maxbytes(inode));
667 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
670 /* We mark all of the fields "set" so MDS/OST does not re-set them */
671 if (attr->ia_valid & ATTR_CTIME) {
672 attr->ia_ctime = CURRENT_TIME;
673 attr->ia_valid |= ATTR_CTIME_SET;
675 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
676 attr->ia_atime = CURRENT_TIME;
677 attr->ia_valid |= ATTR_ATIME_SET;
679 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
680 attr->ia_mtime = CURRENT_TIME;
681 attr->ia_valid |= ATTR_MTIME_SET;
684 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
685 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
686 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
687 LTIME_S(CURRENT_TIME));
689 attr->ia_valid &= ~ATTR_SIZE;
691 /* If only OST attributes being set on objects, don't do MDS RPC.
692 * In that case, we need to check permissions and update the local
693 * inode ourselves so we can call obdo_from_inode() always. */
694 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
696 llu_prepare_mdc_data(&op_data, inode, NULL, NULL, 0, 0);
698 rc = mdc_setattr(sbi->ll_md_exp, &op_data,
699 attr, NULL, 0, NULL, 0, NULL, 0, &request);
702 ptlrpc_req_finished(request);
703 if (rc != -EPERM && rc != -EACCES)
704 CERROR("mdc_setattr fails: rc = %d\n", rc);
708 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
709 sbi->ll_dt_exp, &md);
711 ptlrpc_req_finished(request);
715 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
716 * but needed to set timestamps backwards on utime. */
717 inode_setattr(inode, attr);
718 llu_update_inode(inode, md.body, md.lsm);
719 ptlrpc_req_finished(request);
721 if (!md.lsm || !S_ISREG(lli->lli_st_mode)) {
722 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
726 /* The OST doesn't check permissions, but the alternative is
727 * a gratuitous RPC to the MDS. We already rely on the client
728 * to do read/write/truncate permission checks, so is mtime OK?
730 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
731 /* from sys_utime() */
732 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
733 if (current->fsuid != lli->lli_st_uid &&
734 (rc = ll_permission(inode, MAY_WRITE)) != 0)
737 /* from inode_change_ok() */
738 if (current->fsuid != lli->lli_st_uid &&
739 !capable(CAP_FOWNER))
744 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
745 inode_setattr(inode, attr);
748 if (ia_valid & ATTR_SIZE) {
749 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
751 struct lustre_handle lockh = { 0 };
752 int err, ast_flags = 0;
753 /* XXX when we fix the AST intents to pass the discard-range
754 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
756 if (attr->ia_size == 0)
757 ast_flags = LDLM_AST_DISCARD_DATA;
759 rc = llu_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
761 if (rc != ELDLM_OK) {
767 rc = llu_vmtruncate(inode, attr->ia_size);
769 /* unlock now as we don't mind others file lockers racing with
770 * the mds updates below? */
771 err = llu_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
773 CERROR("llu_extent_unlock failed: %d\n", err);
777 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
780 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
781 lli->lli_st_ino, LTIME_S(attr->ia_mtime));
782 oa.o_id = lsm->lsm_object_id;
783 oa.o_valid = OBD_MD_FLID;
784 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
785 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
786 rc = obd_setattr(sbi->ll_dt_exp, &oa, lsm, NULL, NULL);
788 CERROR("obd_setattr fails: rc=%d\n", rc);
793 /* here we simply act as a thin layer to glue it with
794 * llu_setattr_raw(), which is copy from kernel
796 static int llu_iop_setattr(struct pnode *pno,
799 struct intnl_stat *stbuf)
804 liblustre_wait_event(0);
806 LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
807 SETATTR_UID | SETATTR_GID |
808 SETATTR_LEN | SETATTR_MODE)));
809 memset(&iattr, 0, sizeof(iattr));
811 if (mask & SETATTR_MODE) {
812 iattr.ia_mode = stbuf->st_mode;
813 iattr.ia_valid |= ATTR_MODE;
815 if (mask & SETATTR_MTIME) {
816 iattr.ia_mtime = stbuf->st_mtime;
817 iattr.ia_valid |= ATTR_MTIME;
819 if (mask & SETATTR_ATIME) {
820 iattr.ia_atime = stbuf->st_atime;
821 iattr.ia_valid |= ATTR_ATIME;
823 if (mask & SETATTR_UID) {
824 iattr.ia_uid = stbuf->st_uid;
825 iattr.ia_valid |= ATTR_UID;
827 if (mask & SETATTR_GID) {
828 iattr.ia_gid = stbuf->st_gid;
829 iattr.ia_valid |= ATTR_GID;
831 if (mask & SETATTR_LEN) {
832 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
833 iattr.ia_valid |= ATTR_SIZE;
836 iattr.ia_valid |= ATTR_RAW;
838 RETURN(llu_setattr_raw(ino, &iattr));
841 #define EXT2_LINK_MAX 32000
843 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
845 struct inode *dir = pno->p_base->pb_parent->pb_ino;
846 struct qstr *qstr = &pno->p_base->pb_name;
847 const char *name = qstr->name;
849 struct ptlrpc_request *request = NULL;
850 struct llu_sb_info *sbi = llu_i2sbi(dir);
851 struct mdc_op_data op_data;
855 if (llu_i2info(dir)->lli_st_nlink >= EXT2_LINK_MAX)
858 llu_prepare_mdc_data(&op_data, dir, NULL, name, len, 0);
859 err = mdc_create(sbi->ll_md_exp, &op_data,
860 tgt, strlen(tgt) + 1, S_IFLNK | S_IRWXUGO,
861 current->fsuid, current->fsgid, 0, &request);
862 ptlrpc_req_finished(request);
866 static int llu_readlink_internal(struct inode *inode,
867 struct ptlrpc_request **request,
870 struct llu_inode_info *lli = llu_i2info(inode);
871 struct llu_sb_info *sbi = llu_i2sbi(inode);
873 struct mds_body *body;
874 int rc, symlen = lli->lli_st_size + 1;
879 if (lli->lli_symlink_name) {
880 *symname = lli->lli_symlink_name;
881 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
885 ll_inode2id(&id, inode);
887 /* XXX: capa is NULL here, is it correct? */
888 rc = mdc_getattr(sbi->ll_md_exp, &id, OBD_MD_LINKNAME, NULL, 0,
889 0, symlen, NULL, request);
891 CERROR("inode %lu: rc = %d\n", lli->lli_st_ino, rc);
895 body = lustre_msg_buf ((*request)->rq_repmsg, 0, sizeof (*body));
896 LASSERT (body != NULL);
897 LASSERT_REPSWABBED (*request, 0);
899 if ((body->valid & OBD_MD_LINKNAME) == 0) {
900 CERROR ("OBD_MD_LINKNAME not set on reply\n");
901 GOTO (failed, rc = -EPROTO);
904 LASSERT (symlen != 0);
905 if (body->eadatasize != symlen) {
906 CERROR ("inode %lu: symlink length %d not expected %d\n",
907 lli->lli_st_ino, body->eadatasize - 1, symlen - 1);
908 GOTO (failed, rc = -EPROTO);
911 *symname = lustre_msg_buf ((*request)->rq_repmsg, 1, symlen);
912 if (*symname == NULL ||
913 strnlen (*symname, symlen) != symlen - 1) {
914 /* not full/NULL terminated */
915 CERROR ("inode %lu: symlink not NULL terminated string"
916 "of length %d\n", lli->lli_st_ino, symlen - 1);
917 GOTO (failed, rc = -EPROTO);
920 OBD_ALLOC(lli->lli_symlink_name, symlen);
921 /* do not return an error if we cannot cache the symlink locally */
922 if (lli->lli_symlink_name)
923 memcpy(lli->lli_symlink_name, *symname, symlen);
928 ptlrpc_req_finished (*request);
932 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
934 struct inode *inode = pno->p_base->pb_ino;
935 struct ptlrpc_request *request;
940 rc = llu_readlink_internal(inode, &request, &symname);
945 strncpy(data, symname, bufsize);
947 ptlrpc_req_finished(request);
952 static int llu_iop_mknod_raw(struct pnode *pno,
956 struct ptlrpc_request *request = NULL;
957 struct inode *dir = pno->p_parent->p_base->pb_ino;
958 struct llu_sb_info *sbi = llu_i2sbi(dir);
959 struct mdc_op_data op_data;
963 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu\n",
964 (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
965 llu_i2info(dir)->lli_st_ino);
967 if (llu_i2info(dir)->lli_st_nlink >= EXT2_LINK_MAX)
970 mode &= ~current->fs->umask;
972 switch (mode & S_IFMT) {
975 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
980 llu_prepare_mdc_data(&op_data, dir, NULL,
981 pno->p_base->pb_name.name,
982 pno->p_base->pb_name.len,
984 err = mdc_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
985 current->fsuid, current->fsgid, dev, &request);
986 ptlrpc_req_finished(request);
997 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
999 struct inode *src = old->p_base->pb_ino;
1000 struct inode *dir = new->p_parent->p_base->pb_ino;
1001 const char *name = new->p_base->pb_name.name;
1002 int namelen = new->p_base->pb_name.len;
1003 struct ptlrpc_request *request = NULL;
1004 struct mdc_op_data op_data;
1011 liblustre_wait_event(0);
1012 llu_prepare_mdc_data(&op_data, src, dir, name, namelen, 0);
1013 rc = mdc_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1014 ptlrpc_req_finished(request);
1015 liblustre_wait_event(0);
1021 * libsysio will clear the inode immediately after return
1023 static int llu_iop_unlink_raw(struct pnode *pno)
1025 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1026 struct qstr *qstr = &pno->p_base->pb_name;
1027 const char *name = qstr->name;
1028 int len = qstr->len;
1029 struct inode *target = pno->p_base->pb_ino;
1030 struct ptlrpc_request *request = NULL;
1031 struct mdc_op_data op_data;
1037 liblustre_wait_event(0);
1038 llu_prepare_mdc_data(&op_data, dir, NULL, name, len, 0);
1039 rc = mdc_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1041 rc = llu_objects_destroy(request, dir);
1042 ptlrpc_req_finished(request);
1043 liblustre_wait_event(0);
1047 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1049 struct inode *src = old->p_parent->p_base->pb_ino;
1050 struct inode *tgt = new->p_parent->p_base->pb_ino;
1051 const char *oldname = old->p_base->pb_name.name;
1052 int oldnamelen = old->p_base->pb_name.len;
1053 const char *newname = new->p_base->pb_name.name;
1054 int newnamelen = new->p_base->pb_name.len;
1055 struct ptlrpc_request *request = NULL;
1056 struct mdc_op_data op_data;
1063 llu_prepare_mdc_data(&op_data, src, tgt, NULL, 0, 0);
1064 rc = mdc_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);
1076 #ifdef _HAVE_STATVFS
1077 static int llu_statfs_internal(struct llu_sb_info *sbi,
1078 struct obd_statfs *osfs,
1079 unsigned long max_age)
1081 struct obd_statfs obd_osfs;
1085 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1087 CERROR("mdc_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(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1096 CERROR("obd_statfs fails: rc = %d\n", rc);
1100 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1101 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1104 osfs->os_blocks = obd_osfs.os_blocks;
1105 osfs->os_bfree = obd_osfs.os_bfree;
1106 osfs->os_bavail = obd_osfs.os_bavail;
1108 /* If we don't have as many objects free on the OST as inodes
1109 * on the MDS, we reduce the total number of inodes to
1110 * compensate, so that the "inodes in use" number is correct.
1112 if (obd_osfs.os_ffree < osfs->os_ffree) {
1113 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1115 osfs->os_ffree = obd_osfs.os_ffree;
1121 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1123 struct obd_statfs osfs;
1126 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1128 /* For now we will always get up-to-date statfs values, but in the
1129 * future we may allow some amount of caching on the client (e.g.
1130 * from QOS or lprocfs updates). */
1131 rc = llu_statfs_internal(sbi, &osfs, jiffies - 1);
1135 statfs_unpack(sfs, &osfs);
1137 if (sizeof(sfs->f_blocks) == 4) {
1138 while (osfs.os_blocks > ~0UL) {
1141 osfs.os_blocks >>= 1;
1142 osfs.os_bfree >>= 1;
1143 osfs.os_bavail >>= 1;
1147 sfs->f_blocks = osfs.os_blocks;
1148 sfs->f_bfree = osfs.os_bfree;
1149 sfs->f_bavail = osfs.os_bavail;
1154 static int llu_iop_statvfs(struct pnode *pno,
1156 struct intnl_statvfs *buf)
1162 liblustre_wait_event(0);
1165 LASSERT(pno->p_base->pb_ino);
1166 rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1170 /* from native driver */
1171 buf->f_bsize = fs.f_bsize; /* file system block size */
1172 buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1173 buf->f_blocks = fs.f_blocks;
1174 buf->f_bfree = fs.f_bfree;
1175 buf->f_bavail = fs.f_bavail;
1176 buf->f_files = fs.f_files; /* Total number serial numbers */
1177 buf->f_ffree = fs.f_ffree; /* Number free serial numbers */
1178 buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1179 buf->f_fsid = fs.f_fstc.__val[1];
1180 buf->f_flag = 0; /* No equiv in statfs; maybe use type? */
1181 buf->f_namemax = fs.f_namelen;
1186 #endif /* _HAVE_STATVFS */
1188 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1190 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1191 struct qstr *qstr = &pno->p_base->pb_name;
1192 const char *name = qstr->name;
1193 int len = qstr->len;
1194 struct ptlrpc_request *request = NULL;
1195 struct llu_inode_info *lli = llu_i2info(dir);
1196 struct mdc_op_data op_data;
1199 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%lu(%p)\n",
1200 len, name, lli->lli_st_ino, lli->lli_st_generation, dir);
1202 if (lli->lli_st_nlink >= EXT2_LINK_MAX)
1205 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1206 llu_prepare_mdc_data(&op_data, dir, NULL, name, len, 0);
1207 err = mdc_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0, mode,
1208 current->fsuid, current->fsgid, 0, &request);
1209 ptlrpc_req_finished(request);
1213 static int llu_iop_rmdir_raw(struct pnode *pno)
1215 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1216 struct qstr *qstr = &pno->p_base->pb_name;
1217 const char *name = qstr->name;
1218 int len = qstr->len;
1219 struct ptlrpc_request *request = NULL;
1220 struct mdc_op_data op_data;
1221 struct llu_inode_info *lli = llu_i2info(dir);
1224 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%lu(%p)\n",
1225 len, name, lli->lli_st_ino, lli->lli_st_generation, dir);
1227 llu_prepare_mdc_data(&op_data, dir, NULL, name, len, S_IFDIR);
1228 rc = mdc_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1229 ptlrpc_req_finished(request);
1235 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1237 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1239 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1241 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1243 struct llu_inode_info *lli = llu_i2info(ino);
1248 *rtn = lli->lli_open_flags;
1251 flags = va_arg(ap, long);
1252 flags &= FCNTL_FLMASK;
1253 if (flags & FCNTL_FLMASK_INVALID) {
1254 CERROR("liblustre don't support O_NONBLOCK, O_ASYNC, "
1255 "and O_DIRECT on file descriptor\n");
1259 lli->lli_open_flags = (int) flags;
1264 CERROR("unsupported fcntl cmd %x\n", cmd);
1269 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1271 struct llu_inode_info *lli = llu_i2info(inode);
1272 struct ll_file_data *fd = lli->lli_file_data;
1273 ldlm_policy_data_t policy = { .l_extent = { .start = 0,
1274 .end = OBD_OBJECT_EOF}};
1275 struct lustre_handle lockh = { 0 };
1276 struct lov_stripe_md *lsm = lli->lli_smd;
1281 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1285 policy.l_extent.gid = arg;
1286 if (lli->lli_open_flags & O_NONBLOCK)
1287 flags = LDLM_FL_BLOCK_NOWAIT;
1289 err = llu_extent_lock(fd, inode, lsm, LCK_GROUP, &policy, &lockh,
1294 fd->fd_flags |= LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK;
1296 memcpy(&fd->fd_cwlockh, &lockh, sizeof(lockh));
1301 static int llu_put_grouplock(struct inode *inode, unsigned long arg)
1303 struct llu_inode_info *lli = llu_i2info(inode);
1304 struct ll_file_data *fd = lli->lli_file_data;
1305 struct lov_stripe_md *lsm = lli->lli_smd;
1309 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1312 if (fd->fd_gid != arg)
1315 fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK);
1317 err = llu_extent_unlock(fd, inode, lsm, LCK_GROUP, &fd->fd_cwlockh);
1322 memset(&fd->fd_cwlockh, 0, sizeof(fd->fd_cwlockh));
1327 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1332 liblustre_wait_event(0);
1335 case LL_IOC_GROUP_LOCK:
1336 arg = va_arg(ap, unsigned long);
1337 return llu_get_grouplock(ino, arg);
1338 case LL_IOC_GROUP_UNLOCK:
1339 arg = va_arg(ap, unsigned long);
1340 return llu_put_grouplock(ino, arg);
1343 CERROR("did not support ioctl cmd %lx\n", request);
1348 * we already do syncronous read/write
1350 static int llu_iop_sync(struct inode *inode)
1352 liblustre_wait_event(0);
1356 static int llu_iop_datasync(struct inode *inode)
1358 liblustre_wait_event(0);
1362 struct filesys_ops llu_filesys_ops =
1364 fsop_gone: llu_fsop_gone,
1367 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1369 struct inode *inode;
1370 struct lustre_id id;
1371 struct file_identifier fileid = {&id, sizeof(id)};
1373 if ((md->body->valid &
1374 (OBD_MD_FLGENER | OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1375 (OBD_MD_FLGENER | OBD_MD_FLID | OBD_MD_FLTYPE)) {
1376 CERROR("bad md body valid mask 0x"LPX64"\n",
1379 return ERR_PTR(-EPERM);
1384 /* try to find existing inode */
1385 inode = _sysio_i_find(fs, &fileid);
1387 struct llu_inode_info *lli = llu_i2info(inode);
1389 if (inode->i_zombie ||
1390 lli->lli_st_generation != id_gen(&md->body->id1)) {
1394 llu_update_inode(inode, md->body, md->lsm);
1399 inode = llu_new_inode(fs, &id);
1401 llu_update_inode(inode, md->body, md->lsm);
1407 llu_fsswop_mount(const char *source,
1409 const void *data __IS_UNUSED,
1410 struct pnode *tocover,
1411 struct mount **mntp)
1415 struct pnode_base *rootpb;
1416 struct obd_device *obd;
1417 struct lustre_id rootid;
1418 struct llu_sb_info *sbi;
1419 struct obd_statfs osfs;
1420 static struct qstr noname = { NULL, 0, 0 };
1421 struct ptlrpc_request *request = NULL;
1422 struct lustre_handle lmv_conn = {0, };
1423 struct lustre_handle lov_conn = {0, };
1424 struct lustre_md md;
1426 struct config_llog_instance cfg;
1427 struct lustre_profile *lprof;
1428 char *lov = NULL, *lmv = NULL;
1429 int async = 1, err = -EINVAL;
1433 /* allocate & initialize sbi */
1434 OBD_ALLOC(sbi, sizeof(*sbi));
1438 INIT_LIST_HEAD(&sbi->ll_conn_chain);
1439 generate_random_uuid(uuid);
1440 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1442 /* generate a string unique to this super, let's try
1443 the address of the super itself.*/
1444 OBD_ALLOC(sbi->ll_instance, sizeof(sbi) * 2 + 1);
1445 if (sbi->ll_instance == NULL)
1446 GOTO(out_free, err = -ENOMEM);
1447 sprintf(sbi->ll_instance, "%p", sbi);
1449 /* retrive & parse config log */
1450 cfg.cfg_instance = sbi->ll_instance;
1451 cfg.cfg_uuid = sbi->ll_sb_uuid;
1452 err = liblustre_process_log(&cfg, 1);
1454 CERROR("Unable to process log: %s\n", g_zconf_profile);
1455 GOTO(out_free, err);
1458 lprof = class_get_profile(g_zconf_profile);
1459 if (lprof == NULL) {
1460 CERROR("No profile found: %s\n", g_zconf_profile);
1461 GOTO(out_free, err = -EINVAL);
1464 OBD_FREE(lov, strlen(lov) + 1);
1465 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
1466 strlen(sbi->ll_instance) + 2);
1467 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
1470 OBD_FREE(lmv, strlen(lmv) + 1);
1471 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
1472 strlen(sbi->ll_instance) + 2);
1473 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
1477 GOTO(out_free, err = -EINVAL);
1481 GOTO(out_free, err = -EINVAL);
1484 fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1490 obd = class_name2obd(lmv);
1492 CERROR("MDC %s: not setup or attached\n", lmv);
1493 GOTO(out_free, err = -EINVAL);
1495 obd_set_info(obd->obd_self_export, strlen("async"), "async",
1496 sizeof(async), &async);
1498 if (mdc_init_ea_size(obd, osc))
1499 GOTO(out_free, err = -EINVAL);
1502 err = obd_connect(&lmv_conn, obd, &sbi->ll_sb_uuid, NULL, 0);
1504 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
1505 GOTO(out_free, err);
1507 sbi->ll_md_exp = class_conn2export(&lmv_conn);
1509 err = obd_statfs(obd, &osfs, 100000000);
1514 * FIXME fill fs stat data into sbi here!!! FIXME
1518 obd = class_name2obd(lov);
1520 CERROR("OSC %s: not setup or attached\n", lov);
1521 GOTO(out_lmv, err = -EINVAL);
1523 obd_set_info(obd->obd_self_export, strlen("async"), "async",
1524 sizeof(async), &async);
1526 err = obd_connect(&lov_conn, obd, &sbi->ll_sb_uuid, NULL, 0);
1528 CERROR("cannot connect to %s: rc = %d\n", lov, err);
1531 sbi->ll_dt_exp = class_conn2export(&lov_conn);
1533 err = mdc_getstatus(sbi->ll_md_exp, &rootid);
1535 CERROR("cannot mds_connect: rc = %d\n", err);
1538 CDEBUG(D_SUPER, "rootid "LPU64"\n", rootid.li_stc.u.e3s.l3s_ino);
1539 sbi->ll_rootino = rootid.li_stc.u.e3s.l3s_ino;
1541 /* XXX: capa is NULL here, is it correct? */
1542 err = mdc_getattr(sbi->ll_md_exp, &rootid,
1543 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS), NULL, 0,
1544 0, 0, NULL, &request);
1546 CERROR("mdc_getattr failed for root: rc = %d\n", err);
1550 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1551 sbi->ll_dt_exp, &md);
1553 CERROR("failed to understand root inode md: rc = %d\n",err);
1554 GOTO(out_request, err);
1557 LASSERT(sbi->ll_rootino != 0);
1559 root = llu_iget(fs, &md);
1560 if (!root || IS_ERR(root)) {
1561 CERROR("fail to generate root inode\n");
1562 GOTO(out_request, err = -EBADF);
1566 * Generate base path-node for root.
1568 rootpb = _sysio_pb_new(&noname, NULL, root);
1574 err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
1576 _sysio_pb_gone(rootpb);
1580 ptlrpc_req_finished(request);
1582 printf("LibLustre: namespace mounted successfully!\n");
1587 _sysio_i_gone(root);
1589 ptlrpc_req_finished(request);
1591 obd_disconnect(sbi->ll_dt_exp, 0);
1593 obd_disconnect(sbi->ll_md_exp, 0);
1595 OBD_FREE(sbi, sizeof(*sbi));
1599 struct fssw_ops llu_fssw_ops = {
1603 struct inode_ops llu_inode_ops = {
1604 .inop_lookup = llu_iop_lookup,
1605 .inop_getattr = llu_iop_getattr,
1606 .inop_setattr = llu_iop_setattr,
1607 .inop_getdirentries = llu_iop_getdirentries,
1608 .inop_mkdir = llu_iop_mkdir_raw,
1609 .inop_rmdir = llu_iop_rmdir_raw,
1610 .inop_symlink = llu_iop_symlink_raw,
1611 .inop_readlink = llu_iop_readlink,
1612 .inop_open = llu_iop_open,
1613 .inop_close = llu_iop_close,
1614 .inop_link = llu_iop_link_raw,
1615 .inop_unlink = llu_iop_unlink_raw,
1616 .inop_rename = llu_iop_rename_raw,
1617 .inop_iodone = llu_iop_iodone,
1618 .inop_fcntl = llu_iop_fcntl,
1619 .inop_sync = llu_iop_sync,
1620 .inop_read = llu_iop_read,
1621 .inop_write = llu_iop_write,
1622 .inop_datasync = llu_iop_datasync,
1623 .inop_ioctl = llu_iop_ioctl,
1624 .inop_mknod = llu_iop_mknod_raw,
1625 #ifdef _HAVE_STATVFS
1626 .inop_statvfs = llu_iop_statvfs,
1628 .inop_gone = llu_iop_gone,