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, 2012, 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 (cfs_curproc_is_in_groups(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);
392 CERROR("REPORT THIS LINE TO PETER\n");
396 if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
398 struct ptlrpc_request *req = NULL;
399 struct llu_sb_info *sbi = llu_i2sbi(inode);
400 struct md_op_data op_data = { { 0 } };
401 unsigned long valid = OBD_MD_FLGETATTR;
404 /* Why don't we update all valid MDS fields here, if we're
405 * doing an RPC anyways? -phil */
406 if (S_ISREG(st->st_mode)) {
407 ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
408 valid |= OBD_MD_FLEASIZE;
411 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen,
413 op_data.op_valid = valid;
415 rc = md_getattr(sbi->ll_md_exp, &op_data, &req);
417 CERROR("failure %d inode %llu\n", rc,
418 (long long)st->st_ino);
421 rc = md_get_lustre_md(sbi->ll_md_exp, req,
422 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
424 /* XXX Too paranoid? */
425 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
426 !((md.body->valid & OBD_MD_FLNLINK) &&
427 (md.body->nlink == 0))) {
428 CERROR("Asked for %s eadata but got %s (%d)\n",
429 (valid & OBD_MD_FLEASIZE) ? "some" : "no",
430 (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
431 md.body->eadatasize);
434 ptlrpc_req_finished(req);
439 llu_update_inode(inode, &md);
441 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
442 ptlrpc_req_finished(req);
445 if (!lli->lli_has_smd) {
446 /* object not yet allocated, don't validate size */
447 st->st_atime = lli->lli_lvb.lvb_atime;
448 st->st_mtime = lli->lli_lvb.lvb_mtime;
449 st->st_ctime = lli->lli_lvb.lvb_ctime;
453 /* ll_glimpse_size will prefer locally cached writes if they extend
455 RETURN(cl_glimpse_size(inode));
458 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
460 *b = *llu_i2stat(ino);
463 static int llu_iop_getattr(struct pnode *pno,
465 struct intnl_stat *b)
470 liblustre_wait_event(0);
474 LASSERT(pno->p_base->pb_ino);
475 ino = pno->p_base->pb_ino;
477 LASSERT(!pno || pno->p_base->pb_ino == ino);
480 /* libsysio might call us directly without intent lock,
481 * we must re-fetch the attrs here
483 rc = llu_inode_revalidate(ino);
485 copy_stat_buf(ino, b);
486 LASSERT(!llu_i2info(ino)->lli_it);
489 liblustre_wait_event(0);
493 static int null_if_equal(struct ldlm_lock *lock, void *data)
495 if (data == lock->l_ast_data) {
496 lock->l_ast_data = NULL;
498 if (lock->l_req_mode != lock->l_granted_mode)
499 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
502 return LDLM_ITER_CONTINUE;
505 void llu_clear_inode(struct inode *inode)
507 struct llu_inode_info *lli = llu_i2info(inode);
508 struct llu_sb_info *sbi = llu_i2sbi(inode);
509 struct lov_stripe_md *lsm;
512 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
513 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
516 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
517 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
519 lsm = ccc_inode_lsm_get(inode);
521 obd_change_cbdata(sbi->ll_dt_exp, lsm, null_if_equal, inode);
522 ccc_inode_lsm_put(inode, lsm);
524 cl_inode_fini(inode);
525 lli->lli_has_smd = false;
527 if (lli->lli_symlink_name) {
528 OBD_FREE(lli->lli_symlink_name,
529 strlen(lli->lli_symlink_name) + 1);
530 lli->lli_symlink_name = NULL;
536 void llu_iop_gone(struct inode *inode)
538 struct llu_inode_info *lli = llu_i2info(inode);
541 liblustre_wait_event(0);
542 llu_clear_inode(inode);
544 OBD_FREE(lli, sizeof(*lli));
548 static int inode_setattr(struct inode * inode, struct iattr * attr)
550 unsigned int ia_valid = attr->ia_valid;
551 struct intnl_stat *st = llu_i2stat(inode);
555 * inode_setattr() is only ever invoked with ATTR_SIZE (by
556 * llu_setattr_raw()) when file has no bodies. Check this.
558 LASSERT(ergo(ia_valid & ATTR_SIZE, !llu_i2info(inode)->lli_has_smd));
560 if (ia_valid & ATTR_SIZE)
561 st->st_size = attr->ia_size;
562 if (ia_valid & ATTR_UID)
563 st->st_uid = attr->ia_uid;
564 if (ia_valid & ATTR_GID)
565 st->st_gid = attr->ia_gid;
566 if (ia_valid & ATTR_ATIME)
567 st->st_atime = attr->ia_atime;
568 if (ia_valid & ATTR_MTIME)
569 st->st_mtime = attr->ia_mtime;
570 if (ia_valid & ATTR_CTIME)
571 st->st_ctime = attr->ia_ctime;
572 if (ia_valid & ATTR_MODE) {
573 st->st_mode = attr->ia_mode;
574 if (!cfs_curproc_is_in_groups(st->st_gid) &&
575 !cfs_capable(CFS_CAP_FSETID))
576 st->st_mode &= ~S_ISGID;
578 /* mark_inode_dirty(inode); */
582 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
583 struct md_open_data **mod)
586 struct llu_sb_info *sbi = llu_i2sbi(inode);
587 struct ptlrpc_request *request = NULL;
591 llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
592 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
596 ptlrpc_req_finished(request);
597 if (rc != -EPERM && rc != -EACCES)
598 CERROR("md_setattr fails: rc = %d\n", rc);
602 rc = md_get_lustre_md(sbi->ll_md_exp, request,
603 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
605 ptlrpc_req_finished(request);
609 /* We call inode_setattr to adjust timestamps.
610 * If there is at least some data in file, we cleared ATTR_SIZE
611 * above to avoid invoking vmtruncate, otherwise it is important
612 * to call vmtruncate in inode_setattr to update inode->i_size
614 inode_setattr(inode, &op_data->op_attr);
615 llu_update_inode(inode, &md);
616 ptlrpc_req_finished(request);
621 /* Close IO epoch and send Size-on-MDS attribute update. */
622 static int llu_setattr_done_writing(struct inode *inode,
623 struct md_op_data *op_data,
624 struct md_open_data *mod)
626 struct llu_inode_info *lli = llu_i2info(inode);
627 struct intnl_stat *st = llu_i2stat(inode);
631 LASSERT(op_data != NULL);
632 if (!S_ISREG(st->st_mode))
635 /* XXX: pass och here for the recovery purpose. */
636 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
637 op_data->op_ioepoch, PFID(&lli->lli_fid));
639 op_data->op_flags = MF_EPOCH_CLOSE;
640 llu_done_writing_attr(inode, op_data);
641 llu_pack_inode2opdata(inode, op_data, NULL);
643 rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
645 /* MDS has instructed us to obtain Size-on-MDS attribute
646 * from OSTs and send setattr to back to MDS. */
647 rc = llu_som_update(inode, op_data);
649 CERROR("inode %llu mdc truncate failed: rc = %d\n",
650 (unsigned long long)st->st_ino, rc);
655 /* If this inode has objects allocated to it (lsm != NULL), then the OST
656 * object(s) determine the file size and mtime. Otherwise, the MDS will
657 * keep these values until such a time that objects are allocated for it.
658 * We do the MDS operations first, as it is checking permissions for us.
659 * We don't to the MDS RPC if there is nothing that we want to store there,
660 * otherwise there is no harm in updating mtime/atime on the MDS if we are
661 * going to do an RPC anyways.
663 * If we are doing a truncate, we will send the mtime and ctime updates
664 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
665 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
668 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
670 int has_lsm = llu_i2info(inode)->lli_has_smd;
671 struct intnl_stat *st = llu_i2stat(inode);
672 int ia_valid = attr->ia_valid;
673 struct md_op_data op_data = { { 0 } };
674 struct md_open_data *mod = NULL;
678 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
680 if (ia_valid & ATTR_SIZE) {
681 if (attr->ia_size > ll_file_maxbytes(inode)) {
682 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
683 (long long)attr->ia_size,
684 ll_file_maxbytes(inode));
688 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
691 /* We mark all of the fields "set" so MDS/OST does not re-set them */
692 if (attr->ia_valid & ATTR_CTIME) {
693 attr->ia_ctime = CFS_CURRENT_TIME;
694 attr->ia_valid |= ATTR_CTIME_SET;
696 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
697 attr->ia_atime = CFS_CURRENT_TIME;
698 attr->ia_valid |= ATTR_ATIME_SET;
700 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
701 attr->ia_mtime = CFS_CURRENT_TIME;
702 attr->ia_valid |= ATTR_MTIME_SET;
705 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
706 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
707 ", now = "CFS_TIME_T"\n",
708 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
709 LTIME_S(CFS_CURRENT_TIME));
711 /* NB: ATTR_SIZE will only be set after this point if the size
712 * resides on the MDS, ie, this file has no objects. */
714 attr->ia_valid &= ~ATTR_SIZE;
716 /* If only OST attributes being set on objects, don't do MDS RPC.
717 * In that case, we need to check permissions and update the local
718 * inode ourselves so we can call obdo_from_inode() always. */
719 if (ia_valid & (has_lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
720 memcpy(&op_data.op_attr, attr, sizeof(*attr));
722 /* Open epoch for truncate. */
723 if (exp_connect_som(llu_i2mdexp(inode)) &&
724 (ia_valid & ATTR_SIZE))
725 op_data.op_flags = MF_EPOCH_OPEN;
726 rc = llu_md_setattr(inode, &op_data, &mod);
730 llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch);
731 if (!has_lsm || !S_ISREG(st->st_mode)) {
732 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
736 /* The OST doesn't check permissions, but the alternative is
737 * a gratuitous RPC to the MDS. We already rely on the client
738 * to do read/write/truncate permission checks, so is mtime OK?
740 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
741 /* from sys_utime() */
742 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
743 if (current->fsuid != st->st_uid &&
744 (rc = ll_permission(inode, MAY_WRITE)) != 0)
747 /* from inode_change_ok() */
748 if (current->fsuid != st->st_uid &&
749 !cfs_capable(CFS_CAP_FOWNER))
755 /* Won't invoke llu_vmtruncate(), as we already cleared
757 inode_setattr(inode, attr);
760 if (ia_valid & ATTR_SIZE)
761 attr->ia_valid |= ATTR_SIZE;
762 if (ia_valid & (ATTR_SIZE |
763 ATTR_ATIME | ATTR_ATIME_SET |
764 ATTR_MTIME | ATTR_MTIME_SET))
765 /* on truncate and utimes send attributes to osts, setting
766 * mtime/atime to past will be performed under PW 0:EOF extent
767 * lock (new_size:EOF for truncate)
768 * it may seem excessive to send mtime/atime updates to osts
769 * when not setting times to past, but it is necessary due to
770 * possible time de-synchronization */
771 rc = cl_setattr_ost(inode, attr, NULL);
774 if (op_data.op_ioepoch)
775 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
776 return rc ? rc : rc1;
779 /* here we simply act as a thin layer to glue it with
780 * llu_setattr_raw(), which is copy from kernel
782 static int llu_iop_setattr(struct pnode *pno,
785 struct intnl_stat *stbuf)
791 liblustre_wait_event(0);
793 LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
794 SETATTR_UID | SETATTR_GID |
795 SETATTR_LEN | SETATTR_MODE)));
796 memset(&iattr, 0, sizeof(iattr));
798 if (mask & SETATTR_MODE) {
799 iattr.ia_mode = stbuf->st_mode;
800 iattr.ia_valid |= ATTR_MODE;
802 if (mask & SETATTR_MTIME) {
803 iattr.ia_mtime = stbuf->st_mtime;
804 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
806 if (mask & SETATTR_ATIME) {
807 iattr.ia_atime = stbuf->st_atime;
808 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
810 if (mask & SETATTR_UID) {
811 iattr.ia_uid = stbuf->st_uid;
812 iattr.ia_valid |= ATTR_UID;
814 if (mask & SETATTR_GID) {
815 iattr.ia_gid = stbuf->st_gid;
816 iattr.ia_valid |= ATTR_GID;
818 if (mask & SETATTR_LEN) {
819 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
820 iattr.ia_valid |= ATTR_SIZE;
823 iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
824 iattr.ia_ctime = CFS_CURRENT_TIME;
826 rc = llu_setattr_raw(ino, &iattr);
827 liblustre_wait_idle();
831 #define EXT2_LINK_MAX 32000
833 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
835 struct inode *dir = pno->p_base->pb_parent->pb_ino;
836 struct qstr *qstr = &pno->p_base->pb_name;
837 const char *name = qstr->name;
839 struct ptlrpc_request *request = NULL;
840 struct llu_sb_info *sbi = llu_i2sbi(dir);
841 struct md_op_data op_data = {{ 0 }};
845 liblustre_wait_event(0);
846 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
849 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
852 err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
853 S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
854 cfs_curproc_cap_pack(), 0, &request);
855 ptlrpc_req_finished(request);
856 liblustre_wait_event(0);
860 static int llu_readlink_internal(struct inode *inode,
861 struct ptlrpc_request **request,
864 struct llu_inode_info *lli = llu_i2info(inode);
865 struct llu_sb_info *sbi = llu_i2sbi(inode);
866 struct mdt_body *body;
867 struct intnl_stat *st = llu_i2stat(inode);
868 struct md_op_data op_data = {{ 0 }};
869 int rc, symlen = st->st_size + 1;
875 if (lli->lli_symlink_name) {
876 *symname = lli->lli_symlink_name;
877 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
881 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen,
883 op_data.op_valid = OBD_MD_LINKNAME;
885 rc = md_getattr(sbi->ll_md_exp, &op_data, request);
887 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
891 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
892 LASSERT(body != NULL);
894 if ((body->valid & OBD_MD_LINKNAME) == 0) {
895 CERROR ("OBD_MD_LINKNAME not set on reply\n");
896 GOTO (failed, rc = -EPROTO);
899 LASSERT(symlen != 0);
900 if (body->eadatasize != symlen) {
901 CERROR("inode %llu: symlink length %d not expected %d\n",
902 (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
903 GOTO(failed, rc = -EPROTO);
906 *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
907 if (*symname == NULL ||
908 strnlen(*symname, symlen) != symlen - 1) {
909 /* not full/NULL terminated */
910 CERROR("inode %llu: symlink not NULL terminated string"
911 "of length %d\n", (long long)st->st_ino, symlen - 1);
912 GOTO(failed, rc = -EPROTO);
915 OBD_ALLOC(lli->lli_symlink_name, symlen);
916 /* do not return an error if we cannot cache the symlink locally */
917 if (lli->lli_symlink_name)
918 memcpy(lli->lli_symlink_name, *symname, symlen);
923 ptlrpc_req_finished (*request);
927 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
929 struct inode *inode = pno->p_base->pb_ino;
930 struct ptlrpc_request *request;
935 liblustre_wait_event(0);
936 rc = llu_readlink_internal(inode, &request, &symname);
941 strncpy(data, symname, bufsize);
942 rc = strlen(symname);
944 ptlrpc_req_finished(request);
946 liblustre_wait_event(0);
950 static int llu_iop_mknod_raw(struct pnode *pno,
954 struct ptlrpc_request *request = NULL;
955 struct inode *dir = pno->p_parent->p_base->pb_ino;
956 struct llu_sb_info *sbi = llu_i2sbi(dir);
957 struct md_op_data op_data = {{ 0 }};
961 liblustre_wait_event(0);
962 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
963 (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
964 (long long)llu_i2stat(dir)->st_ino);
966 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
969 switch (mode & S_IFMT) {
972 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
977 llu_prep_md_op_data(&op_data, dir, NULL,
978 pno->p_base->pb_name.name,
979 pno->p_base->pb_name.len, 0,
982 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
983 current->fsuid, current->fsgid,
984 cfs_curproc_cap_pack(), dev, &request);
985 ptlrpc_req_finished(request);
993 liblustre_wait_event(0);
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 md_op_data op_data = {{ 0 }};
1011 liblustre_wait_event(0);
1012 llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1014 rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1015 ptlrpc_req_finished(request);
1016 liblustre_wait_event(0);
1022 * libsysio will clear the inode immediately after return
1024 static int llu_iop_unlink_raw(struct pnode *pno)
1026 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1027 struct qstr *qstr = &pno->p_base->pb_name;
1028 const char *name = qstr->name;
1029 int len = qstr->len;
1030 struct inode *target = pno->p_base->pb_ino;
1031 struct ptlrpc_request *request = NULL;
1032 struct md_op_data op_data = { { 0 } };
1038 liblustre_wait_event(0);
1039 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1041 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1043 rc = llu_objects_destroy(request, dir);
1044 ptlrpc_req_finished(request);
1045 liblustre_wait_idle();
1050 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1052 struct inode *src = old->p_parent->p_base->pb_ino;
1053 struct inode *tgt = new->p_parent->p_base->pb_ino;
1054 const char *oldname = old->p_base->pb_name.name;
1055 int oldnamelen = old->p_base->pb_name.len;
1056 const char *newname = new->p_base->pb_name.name;
1057 int newnamelen = new->p_base->pb_name.len;
1058 struct ptlrpc_request *request = NULL;
1059 struct md_op_data op_data = { { 0 } };
1066 liblustre_wait_event(0);
1067 llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1069 rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1070 oldname, oldnamelen, newname, newnamelen,
1073 rc = llu_objects_destroy(request, src);
1076 ptlrpc_req_finished(request);
1077 liblustre_wait_idle();
1082 #ifdef _HAVE_STATVFS
1083 static int llu_statfs_internal(struct llu_sb_info *sbi,
1084 struct obd_statfs *osfs, __u64 max_age)
1086 struct obd_statfs obd_osfs;
1090 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, 0);
1092 CERROR("md_statfs fails: rc = %d\n", rc);
1096 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1097 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1099 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1100 &obd_statfs, max_age, 0);
1102 CERROR("obd_statfs fails: rc = %d\n", rc);
1106 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1107 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1110 osfs->os_blocks = obd_osfs.os_blocks;
1111 osfs->os_bfree = obd_osfs.os_bfree;
1112 osfs->os_bavail = obd_osfs.os_bavail;
1114 /* If we don't have as many objects free on the OST as inodes
1115 * on the MDS, we reduce the total number of inodes to
1116 * compensate, so that the "inodes in use" number is correct.
1118 if (obd_osfs.os_ffree < osfs->os_ffree) {
1119 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1121 osfs->os_ffree = obd_osfs.os_ffree;
1127 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1129 struct obd_statfs osfs;
1132 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1134 /* For now we will always get up-to-date statfs values, but in the
1135 * future we may allow some amount of caching on the client (e.g.
1136 * from QOS or lprocfs updates). */
1137 rc = llu_statfs_internal(sbi, &osfs,
1138 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS));
1142 statfs_unpack(sfs, &osfs);
1144 if (sizeof(sfs->f_blocks) == 4) {
1145 while (osfs.os_blocks > ~0UL) {
1148 osfs.os_blocks >>= 1;
1149 osfs.os_bfree >>= 1;
1150 osfs.os_bavail >>= 1;
1154 sfs->f_blocks = osfs.os_blocks;
1155 sfs->f_bfree = osfs.os_bfree;
1156 sfs->f_bavail = osfs.os_bavail;
1161 static int llu_iop_statvfs(struct pnode *pno,
1163 struct intnl_statvfs *buf)
1169 liblustre_wait_event(0);
1172 LASSERT(pno->p_base->pb_ino);
1173 rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1177 /* from native driver */
1178 buf->f_bsize = fs.f_bsize; /* file system block size */
1179 buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1180 buf->f_blocks = fs.f_blocks;
1181 buf->f_bfree = fs.f_bfree;
1182 buf->f_bavail = fs.f_bavail;
1183 buf->f_files = fs.f_files; /* Total number serial numbers */
1184 buf->f_ffree = fs.f_ffree; /* Number free serial numbers */
1185 buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1186 buf->f_fsid = fs.f_fsid.__val[1];
1187 buf->f_flag = 0; /* No equiv in statfs; maybe use type? */
1188 buf->f_namemax = fs.f_namelen;
1191 liblustre_wait_event(0);
1194 #endif /* _HAVE_STATVFS */
1196 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1198 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1199 struct qstr *qstr = &pno->p_base->pb_name;
1200 const char *name = qstr->name;
1201 int len = qstr->len;
1202 struct ptlrpc_request *request = NULL;
1203 struct intnl_stat *st = llu_i2stat(dir);
1204 struct md_op_data op_data = {{ 0 }};
1208 liblustre_wait_event(0);
1209 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1210 (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1212 if (st->st_nlink >= EXT2_LINK_MAX)
1215 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1218 err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1219 mode | S_IFDIR, current->fsuid, current->fsgid,
1220 cfs_curproc_cap_pack(), 0, &request);
1221 ptlrpc_req_finished(request);
1222 liblustre_wait_event(0);
1226 static int llu_iop_rmdir_raw(struct pnode *pno)
1228 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1229 struct qstr *qstr = &pno->p_base->pb_name;
1230 const char *name = qstr->name;
1231 int len = qstr->len;
1232 struct ptlrpc_request *request = NULL;
1233 struct md_op_data op_data = {{ 0 }};
1237 liblustre_wait_event(0);
1238 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1239 (long long)llu_i2stat(dir)->st_ino,
1240 llu_i2info(dir)->lli_st_generation, dir);
1242 llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1244 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1245 ptlrpc_req_finished(request);
1247 liblustre_wait_event(0);
1252 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1254 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1256 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1258 /* refer to ll_file_flock() for details */
1259 static int llu_file_flock(struct inode *ino,
1261 struct file_lock *file_lock)
1263 struct llu_inode_info *lli = llu_i2info(ino);
1264 struct intnl_stat *st = llu_i2stat(ino);
1265 struct ldlm_res_id res_id =
1266 { .name = {fid_seq(&lli->lli_fid),
1267 fid_oid(&lli->lli_fid),
1268 fid_ver(&lli->lli_fid),
1270 struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
1271 ldlm_flock_completion_ast, NULL, NULL, file_lock };
1273 struct lustre_handle lockh = {0};
1274 ldlm_policy_data_t flock;
1278 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1279 (unsigned long long)st->st_ino, file_lock);
1281 flock.l_flock.pid = file_lock->fl_pid;
1282 flock.l_flock.start = file_lock->fl_start;
1283 flock.l_flock.end = file_lock->fl_end;
1285 switch (file_lock->fl_type) {
1287 einfo.ei_mode = LCK_PR;
1290 einfo.ei_mode = LCK_NL;
1293 einfo.ei_mode = LCK_PW;
1296 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1303 #if F_SETLKW64 != F_SETLKW
1311 #if F_SETLK64 != F_SETLK
1315 flags = LDLM_FL_BLOCK_NOWAIT;
1319 #if F_GETLK64 != F_GETLK
1323 flags = LDLM_FL_TEST_LOCK;
1324 file_lock->fl_type = einfo.ei_mode;
1327 CERROR("unknown fcntl cmd: %d\n", cmd);
1331 CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#llx, mode=%u, "
1332 "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1333 flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1337 struct lmv_obd *lmv;
1338 struct obd_device *lmv_obd;
1339 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1340 lmv = &lmv_obd->u.lmv;
1342 if (lmv->desc.ld_tgt_count < 1)
1343 RETURN(rc = -ENODEV);
1345 if (lmv->tgts[0] && lmv->tgts[0]->ltd_exp != NULL)
1346 rc = ldlm_cli_enqueue(lmv->tgts[0]->ltd_exp, NULL,
1347 &einfo, &res_id, &flock, &flags,
1348 NULL, 0, LVB_T_NONE, &lockh, 0);
1355 static int assign_type(struct file_lock *fl, int type)
1368 static int flock_to_posix_lock(struct inode *ino,
1369 struct file_lock *fl,
1372 switch (l->l_whence) {
1373 /* XXX: only SEEK_SET is supported in lustre */
1381 fl->fl_end = l->l_len - 1;
1385 fl->fl_end = OFFSET_MAX;
1387 fl->fl_pid = getpid();
1388 fl->fl_flags = FL_POSIX;
1389 fl->fl_notify = NULL;
1390 fl->fl_insert = NULL;
1391 fl->fl_remove = NULL;
1392 /* XXX: these fields can't be filled with suitable values,
1393 but I think lustre doesn't use them.
1395 fl->fl_owner = NULL;
1398 return assign_type(fl, l->l_type);
1401 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1403 struct file_lock fl;
1407 if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1410 error = flock_to_posix_lock(ino, &fl, flock);
1414 error = llu_file_flock(ino, F_GETLK, &fl);
1418 flock->l_type = F_UNLCK;
1419 if (fl.fl_type != F_UNLCK) {
1420 flock->l_pid = fl.fl_pid;
1421 flock->l_start = fl.fl_start;
1422 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1423 fl.fl_end - fl.fl_start + 1;
1424 flock->l_whence = SEEK_SET;
1425 flock->l_type = fl.fl_type;
1432 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1434 struct file_lock fl;
1435 int flags = llu_i2info(ino)->lli_open_flags + 1;
1438 error = flock_to_posix_lock(ino, &fl, flock);
1441 if (cmd == F_SETLKW)
1442 fl.fl_flags |= FL_SLEEP;
1445 switch (flock->l_type) {
1447 if (!(flags & FMODE_READ))
1451 if (!(flags & FMODE_WRITE))
1461 error = llu_file_flock(ino, cmd, &fl);
1469 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1471 struct llu_inode_info *lli = llu_i2info(ino);
1473 struct flock *flock;
1476 liblustre_wait_event(0);
1479 *rtn = lli->lli_open_flags;
1482 flags = va_arg(ap, long);
1483 flags &= FCNTL_FLMASK;
1484 if (flags & FCNTL_FLMASK_INVALID) {
1485 LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1486 "the O_NONBLOCK or O_ASYNC flags. "
1487 "Please fix your application.\n");
1492 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1493 (lli->lli_open_flags & ~FCNTL_FLMASK);
1498 #if F_GETLK64 != F_GETLK
1502 flock = va_arg(ap, struct flock *);
1503 err = llu_fcntl_getlk(ino, flock);
1508 #if F_SETLKW64 != F_SETLKW
1514 #if F_SETLK64 != F_SETLK
1518 flock = va_arg(ap, struct flock *);
1519 err = llu_fcntl_setlk(ino, cmd, flock);
1523 CERROR("unsupported fcntl cmd %x\n", cmd);
1529 liblustre_wait_event(0);
1533 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1535 struct llu_inode_info *lli = llu_i2info(inode);
1536 struct ll_file_data *fd = lli->lli_file_data;
1538 struct ccc_grouplock grouplock;
1541 if (fd->fd_flags & LL_FILE_IGNORE_LOCK) {
1544 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1547 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1549 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1550 arg, (lli->lli_open_flags & O_NONBLOCK),
1556 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1557 fd->fd_grouplock = grouplock;
1562 int llu_put_grouplock(struct inode *inode, unsigned long arg)
1564 struct llu_inode_info *lli = llu_i2info(inode);
1565 struct ll_file_data *fd = lli->lli_file_data;
1566 struct ccc_grouplock grouplock;
1569 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1572 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1574 if (fd->fd_grouplock.cg_gid != arg)
1577 grouplock = fd->fd_grouplock;
1578 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1579 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1581 cl_put_grouplock(&grouplock);
1586 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1588 struct llu_sb_info *sbi = llu_i2sbi(ino);
1589 struct ptlrpc_request *request = NULL;
1590 struct md_op_data op_data = {{ 0 }};
1591 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1594 llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1597 LASSERT(sizeof(lum) == sizeof(*lump));
1598 LASSERT(sizeof(lum.lmm_objects[0]) ==
1599 sizeof(lump->lmm_objects[0]));
1600 if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1603 switch (lum.lmm_magic) {
1604 case LOV_USER_MAGIC_V1: {
1605 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1606 lustre_swab_lov_user_md_v1(&lum);
1609 case LOV_USER_MAGIC_V3: {
1610 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1611 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1615 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1616 " %#08x != %#08x nor %#08x\n",
1617 lum.lmm_magic, LOV_USER_MAGIC_V1,
1623 /* swabbing is done in lov_setstripe() on server side */
1624 rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1625 sizeof(lum), NULL, 0, &request, NULL);
1627 ptlrpc_req_finished(request);
1628 if (rc != -EPERM && rc != -EACCES)
1629 CERROR("md_setattr fails: rc = %d\n", rc);
1632 ptlrpc_req_finished(request);
1637 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1638 struct lov_user_md *lum, int lum_size)
1640 struct llu_sb_info *sbi = llu_i2sbi(ino);
1641 struct llu_inode_info *lli = llu_i2info(ino);
1642 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1643 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR,
1644 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
1645 struct ptlrpc_request *req = NULL;
1646 struct lustre_md md;
1647 struct md_op_data data = {{ 0 }};
1648 struct lustre_handle lockh;
1652 if (lli->lli_has_smd) {
1653 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1654 PFID(&lli->lli_fid));
1658 llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1660 rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1661 &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1665 req = oit.d.lustre.it_data;
1666 rc = it_open_error(DISP_IT_EXECD, &oit);
1672 rc = it_open_error(DISP_OPEN_OPEN, &oit);
1678 rc = md_get_lustre_md(sbi->ll_md_exp, req,
1679 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1683 llu_update_inode(ino, &md);
1684 llu_local_open(lli, &oit);
1685 /* release intent */
1686 if (lustre_handle_is_used(&lockh))
1687 ldlm_lock_decref(&lockh, LCK_CR);
1688 ptlrpc_req_finished(req);
1690 rc = llu_file_release(ino);
1695 ptlrpc_req_finished(req);
1699 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1701 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1703 int flags = FMODE_WRITE;
1706 LASSERT(sizeof(lum) == sizeof(*lump));
1707 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1708 if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1711 rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1715 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1717 struct intnl_stat *st = llu_i2stat(ino);
1718 if (S_ISREG(st->st_mode))
1719 return llu_lov_file_setstripe(ino, arg);
1720 if (S_ISDIR(st->st_mode))
1721 return llu_lov_dir_setstripe(ino, arg);
1726 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1728 struct lov_stripe_md *lsm = NULL;
1731 lsm = ccc_inode_lsm_get(ino);
1733 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1735 ccc_inode_lsm_put(ino, lsm);
1739 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1745 liblustre_wait_event(0);
1748 case LL_IOC_GROUP_LOCK:
1749 arg = va_arg(ap, unsigned long);
1750 rc = llu_get_grouplock(ino, arg);
1752 case LL_IOC_GROUP_UNLOCK:
1753 arg = va_arg(ap, unsigned long);
1754 rc = llu_put_grouplock(ino, arg);
1756 case LL_IOC_LOV_SETSTRIPE:
1757 arg = va_arg(ap, unsigned long);
1758 rc = llu_lov_setstripe(ino, arg);
1760 case LL_IOC_LOV_GETSTRIPE:
1761 arg = va_arg(ap, unsigned long);
1762 rc = llu_lov_getstripe(ino, arg);
1765 CERROR("did not support ioctl cmd %lx\n", request);
1770 liblustre_wait_event(0);
1775 * we already do syncronous read/write
1777 static int llu_iop_sync(struct inode *inode)
1779 liblustre_wait_event(0);
1783 static int llu_iop_datasync(struct inode *inode)
1785 liblustre_wait_event(0);
1789 struct filesys_ops llu_filesys_ops =
1791 fsop_gone: llu_fsop_gone,
1794 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1796 struct inode *inode;
1798 struct file_identifier fileid = {&fid, sizeof(fid)};
1800 if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1801 (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1802 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1804 return ERR_PTR(-EPERM);
1807 /* try to find existing inode */
1808 fid = md->body->fid1;
1810 inode = _sysio_i_find(fs, &fileid);
1812 if (inode->i_zombie/* ||
1813 lli->lli_st_generation != md->body->generation*/) {
1817 llu_update_inode(inode, md);
1822 inode = llu_new_inode(fs, &fid);
1824 llu_update_inode(inode, md);
1830 llu_fsswop_mount(const char *source,
1832 const void *data __IS_UNUSED,
1833 struct pnode *tocover,
1834 struct mount **mntp)
1838 struct pnode_base *rootpb;
1839 struct obd_device *obd;
1840 struct llu_sb_info *sbi;
1841 struct obd_statfs osfs;
1842 static struct qstr noname = { NULL, 0, 0 };
1843 struct ptlrpc_request *request = NULL;
1844 struct lustre_md md;
1846 struct config_llog_instance cfg = {0, };
1847 struct lustre_profile *lprof;
1848 char *zconf_mgsnid, *zconf_profile;
1849 char *osc = NULL, *mdc = NULL;
1850 int async = 1, err = -EINVAL;
1851 struct obd_connect_data ocd = {0,};
1852 struct md_op_data op_data = {{0}};
1853 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1854 const int instlen = sizeof(cfg.cfg_instance) * 2 + 2;
1858 if (ll_parse_mount_target(source,
1861 CERROR("mal-formed target %s\n", source);
1864 if (!zconf_mgsnid || !zconf_profile) {
1865 printf("Liblustre: invalid target %s\n", source);
1868 /* allocate & initialize sbi */
1869 OBD_ALLOC(sbi, sizeof(*sbi));
1873 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
1874 ll_generate_random_uuid(uuid);
1875 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1877 /* generate a string unique to this super, let's try
1878 the address of the super itself.*/
1879 cfg.cfg_instance = sbi;
1881 /* retrive & parse config log */
1882 cfg.cfg_uuid = sbi->ll_sb_uuid;
1883 err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
1885 CERROR("Unable to process log: %s\n", zconf_profile);
1886 GOTO(out_free, err);
1889 lprof = class_get_profile(zconf_profile);
1890 if (lprof == NULL) {
1891 CERROR("No profile found: %s\n", zconf_profile);
1892 GOTO(out_free, err = -EINVAL);
1894 OBD_ALLOC(osc, strlen(lprof->lp_dt) + instlen + 2);
1895 sprintf(osc, "%s-%p", lprof->lp_dt, cfg.cfg_instance);
1897 OBD_ALLOC(mdc, strlen(lprof->lp_md) + instlen + 2);
1898 sprintf(mdc, "%s-%p", lprof->lp_md, cfg.cfg_instance);
1902 GOTO(out_free, err = -EINVAL);
1906 GOTO(out_free, err = -EINVAL);
1909 fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1915 obd = class_name2obd(mdc);
1917 CERROR("MDC %s: not setup or attached\n", mdc);
1918 GOTO(out_free, err = -EINVAL);
1920 obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC),
1921 KEY_ASYNC, sizeof(async), &async, NULL);
1923 ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
1924 OBD_CONNECT_FID | OBD_CONNECT_AT |
1925 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
1926 OBD_CONNECT_LVB_TYPE;
1928 #ifdef LIBLUSTRE_POSIX_ACL
1929 ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
1931 ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
1932 ocd.ocd_version = LUSTRE_VERSION_CODE;
1935 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1937 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
1938 GOTO(out_free, err);
1941 err = obd_statfs(NULL, sbi->ll_md_exp, &osfs, 100000000, 0);
1946 * FIXME fill fs stat data into sbi here!!! FIXME
1950 obd = class_name2obd(osc);
1952 CERROR("OSC %s: not setup or attached\n", osc);
1953 GOTO(out_md, err = -EINVAL);
1955 obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC),
1956 KEY_ASYNC, sizeof(async), &async, NULL);
1958 obd->obd_upcall.onu_owner = &sbi->ll_lco;
1959 obd->obd_upcall.onu_upcall = cl_ocd_update;
1961 ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
1962 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
1963 OBD_CONNECT_FID | OBD_CONNECT_AT |
1964 OBD_CONNECT_FULL20 | OBD_CONNECT_EINPROGRESS |
1965 OBD_CONNECT_LVB_TYPE;
1967 ocd.ocd_version = LUSTRE_VERSION_CODE;
1968 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1970 CERROR("cannot connect to %s: rc = %d\n", osc, err);
1973 sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
1974 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
1975 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
1977 fid_zero(&sbi->ll_root_fid);
1978 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
1980 CERROR("cannot mds_connect: rc = %d\n", err);
1981 GOTO(out_lock_cn_cb, err);
1983 if (!fid_is_sane(&sbi->ll_root_fid)) {
1984 CERROR("Invalid root fid during mount\n");
1985 GOTO(out_lock_cn_cb, err = -EINVAL);
1987 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
1989 op_data.op_fid1 = sbi->ll_root_fid;
1990 op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
1991 /* fetch attr of root inode */
1992 err = md_getattr(sbi->ll_md_exp, &op_data, &request);
1994 CERROR("md_getattr failed for root: rc = %d\n", err);
1995 GOTO(out_lock_cn_cb, err);
1998 err = md_get_lustre_md(sbi->ll_md_exp, request,
1999 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2001 CERROR("failed to understand root inode md: rc = %d\n",err);
2002 GOTO(out_request, err);
2005 LASSERT(fid_is_sane(&sbi->ll_root_fid));
2007 root = llu_iget(fs, &md);
2008 if (!root || IS_ERR(root)) {
2009 CERROR("fail to generate root inode\n");
2010 GOTO(out_request, err = -EBADF);
2014 * Generate base path-node for root.
2016 rootpb = _sysio_pb_new(&noname, NULL, root);
2022 err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2024 _sysio_pb_gone(rootpb);
2030 ptlrpc_req_finished(request);
2032 CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2037 _sysio_i_gone(root);
2039 ptlrpc_req_finished(request);
2041 obd_disconnect(sbi->ll_dt_exp);
2043 obd_disconnect(sbi->ll_md_exp);
2046 OBD_FREE(osc, strlen(lprof->lp_dt) + instlen + 2);
2048 OBD_FREE(mdc, strlen(lprof->lp_md) + instlen + 2);
2049 OBD_FREE(sbi, sizeof(*sbi));
2050 liblustre_wait_idle();
2054 struct fssw_ops llu_fssw_ops = {
2058 static struct inode_ops llu_inode_ops = {
2059 inop_lookup: llu_iop_lookup,
2060 inop_getattr: llu_iop_getattr,
2061 inop_setattr: llu_iop_setattr,
2062 inop_filldirentries: llu_iop_filldirentries,
2063 inop_mkdir: llu_iop_mkdir_raw,
2064 inop_rmdir: llu_iop_rmdir_raw,
2065 inop_symlink: llu_iop_symlink_raw,
2066 inop_readlink: llu_iop_readlink,
2067 inop_open: llu_iop_open,
2068 inop_close: llu_iop_close,
2069 inop_link: llu_iop_link_raw,
2070 inop_unlink: llu_iop_unlink_raw,
2071 inop_rename: llu_iop_rename_raw,
2072 inop_pos: llu_iop_pos,
2073 inop_read: llu_iop_read,
2074 inop_write: llu_iop_write,
2075 inop_iodone: llu_iop_iodone,
2076 inop_fcntl: llu_iop_fcntl,
2077 inop_sync: llu_iop_sync,
2078 inop_datasync: llu_iop_datasync,
2079 inop_ioctl: llu_iop_ioctl,
2080 inop_mknod: llu_iop_mknod_raw,
2081 #ifdef _HAVE_STATVFS
2082 inop_statvfs: llu_iop_statvfs,
2084 inop_gone: llu_iop_gone,