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 "LPU64"/"LPU64", valid %x\n",
226 src->o_id, src->o_seq, valid);
228 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
229 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
232 LTIME_S(st->st_mtime), LTIME_S(st->st_ctime),
233 (long)src->o_mtime, (long)src->o_ctime);
235 if (valid & OBD_MD_FLATIME)
236 LTIME_S(st->st_atime) = src->o_atime;
237 if (valid & OBD_MD_FLMTIME)
238 LTIME_S(st->st_mtime) = src->o_mtime;
239 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
240 LTIME_S(st->st_ctime) = src->o_ctime;
241 if (valid & OBD_MD_FLSIZE)
242 st->st_size = src->o_size;
243 if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
244 st->st_blocks = src->o_blocks;
245 if (valid & OBD_MD_FLBLKSZ)
246 st->st_blksize = src->o_blksize;
247 if (valid & OBD_MD_FLTYPE)
248 st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
249 if (valid & OBD_MD_FLMODE)
250 st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
251 if (valid & OBD_MD_FLUID)
252 st->st_uid = src->o_uid;
253 if (valid & OBD_MD_FLGID)
254 st->st_gid = src->o_gid;
255 if (valid & OBD_MD_FLFLAGS)
256 lli->lli_st_flags = src->o_flags;
260 * Performs the getattr on the inode and updates its fields.
261 * If @sync != 0, perform the getattr under the server-side lock.
263 int llu_inode_getattr(struct inode *inode, struct obdo *obdo,
264 __u64 ioepoch, int sync)
266 struct ptlrpc_request_set *set;
267 struct lov_stripe_md *lsm = NULL;
268 struct obd_info oinfo = { { { 0 } } };
272 lsm = ccc_inode_lsm_get(inode);
277 oinfo.oi_oa->o_oi = lsm->lsm_oi;
278 oinfo.oi_oa->o_mode = S_IFREG;
279 oinfo.oi_oa->o_ioepoch = ioepoch;
280 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
281 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
282 OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
283 OBD_MD_FLCTIME | OBD_MD_FLGROUP |
284 OBD_MD_FLATIME | OBD_MD_FLEPOCH;
285 obdo_set_parent_fid(oinfo.oi_oa, &llu_i2info(inode)->lli_fid);
287 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
288 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
291 set = ptlrpc_prep_set();
293 CERROR ("ENOMEM allocing request set\n");
296 rc = obd_getattr_async(llu_i2obdexp(inode), &oinfo, set);
298 rc = ptlrpc_set_wait(set);
299 ptlrpc_set_destroy(set);
301 ccc_inode_lsm_put(inode, lsm);
305 oinfo.oi_oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
306 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
309 obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
310 CDEBUG(D_INODE, "objid "LPX64" size %llu, blocks %llu, "
311 "blksize %llu\n", oinfo.oi_oa->o_id,
312 (long long unsigned)llu_i2stat(inode)->st_size,
313 (long long unsigned)llu_i2stat(inode)->st_blocks,
314 (long long unsigned)llu_i2stat(inode)->st_blksize);
318 static struct inode* llu_new_inode(struct filesys *fs,
322 struct llu_inode_info *lli;
323 struct intnl_stat st = {
326 #ifndef AUTOMOUNT_FILE_NAME
327 .st_mode = fid->f_type & S_IFMT,
329 .st_mode = fid->f_type /* all of the bits! */
332 /* FIXME: fix this later */
339 OBD_ALLOC(lli, sizeof(*lli));
343 /* initialize lli here */
344 lli->lli_sbi = llu_fs2sbi(fs);
345 lli->lli_has_smd = false;
346 lli->lli_symlink_name = NULL;
348 lli->lli_maxbytes = (__u64)(~0UL);
349 lli->lli_file_data = NULL;
351 lli->lli_sysio_fid.fid_data = &lli->lli_fid;
352 lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
355 /* file identifier is needed by functions like _sysio_i_find() */
356 inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
357 &st, 0, &llu_inode_ops, lli);
360 OBD_FREE(lli, sizeof(*lli));
365 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
367 struct lustre_handle lockh;
368 ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
375 fid = &llu_i2info(inode)->lli_fid;
376 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
378 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
379 if (md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
380 LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
386 static int llu_inode_revalidate(struct inode *inode)
388 struct llu_inode_info *lli = llu_i2info(inode);
389 struct intnl_stat *st = llu_i2stat(inode);
393 CERROR("REPORT THIS LINE TO PETER\n");
397 if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
399 struct ptlrpc_request *req = NULL;
400 struct llu_sb_info *sbi = llu_i2sbi(inode);
401 struct md_op_data op_data = { { 0 } };
402 unsigned long valid = OBD_MD_FLGETATTR;
405 /* Why don't we update all valid MDS fields here, if we're
406 * doing an RPC anyways? -phil */
407 if (S_ISREG(st->st_mode)) {
408 ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
409 valid |= OBD_MD_FLEASIZE;
412 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen,
414 op_data.op_valid = valid;
416 rc = md_getattr(sbi->ll_md_exp, &op_data, &req);
418 CERROR("failure %d inode %llu\n", rc,
419 (long long)st->st_ino);
422 rc = md_get_lustre_md(sbi->ll_md_exp, req,
423 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
425 /* XXX Too paranoid? */
426 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
427 !((md.body->valid & OBD_MD_FLNLINK) &&
428 (md.body->nlink == 0))) {
429 CERROR("Asked for %s eadata but got %s (%d)\n",
430 (valid & OBD_MD_FLEASIZE) ? "some" : "no",
431 (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
432 md.body->eadatasize);
435 ptlrpc_req_finished(req);
440 llu_update_inode(inode, &md);
442 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
443 ptlrpc_req_finished(req);
446 if (!lli->lli_has_smd) {
447 /* object not yet allocated, don't validate size */
448 st->st_atime = lli->lli_lvb.lvb_atime;
449 st->st_mtime = lli->lli_lvb.lvb_mtime;
450 st->st_ctime = lli->lli_lvb.lvb_ctime;
454 /* ll_glimpse_size will prefer locally cached writes if they extend
456 RETURN(cl_glimpse_size(inode));
459 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
461 *b = *llu_i2stat(ino);
464 static int llu_iop_getattr(struct pnode *pno,
466 struct intnl_stat *b)
471 liblustre_wait_event(0);
475 LASSERT(pno->p_base->pb_ino);
476 ino = pno->p_base->pb_ino;
478 LASSERT(!pno || pno->p_base->pb_ino == ino);
481 /* libsysio might call us directly without intent lock,
482 * we must re-fetch the attrs here
484 rc = llu_inode_revalidate(ino);
486 copy_stat_buf(ino, b);
487 LASSERT(!llu_i2info(ino)->lli_it);
490 liblustre_wait_event(0);
494 static int null_if_equal(struct ldlm_lock *lock, void *data)
496 if (data == lock->l_ast_data) {
497 lock->l_ast_data = NULL;
499 if (lock->l_req_mode != lock->l_granted_mode)
500 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
503 return LDLM_ITER_CONTINUE;
506 void llu_clear_inode(struct inode *inode)
508 struct llu_inode_info *lli = llu_i2info(inode);
509 struct llu_sb_info *sbi = llu_i2sbi(inode);
510 struct lov_stripe_md *lsm;
513 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
514 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
517 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
518 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
520 lsm = ccc_inode_lsm_get(inode);
522 obd_change_cbdata(sbi->ll_dt_exp, lsm, null_if_equal, inode);
523 ccc_inode_lsm_put(inode, lsm);
525 cl_inode_fini(inode);
526 lli->lli_has_smd = false;
528 if (lli->lli_symlink_name) {
529 OBD_FREE(lli->lli_symlink_name,
530 strlen(lli->lli_symlink_name) + 1);
531 lli->lli_symlink_name = NULL;
537 void llu_iop_gone(struct inode *inode)
539 struct llu_inode_info *lli = llu_i2info(inode);
542 liblustre_wait_event(0);
543 llu_clear_inode(inode);
545 OBD_FREE(lli, sizeof(*lli));
549 static int inode_setattr(struct inode * inode, struct iattr * attr)
551 unsigned int ia_valid = attr->ia_valid;
552 struct intnl_stat *st = llu_i2stat(inode);
556 * inode_setattr() is only ever invoked with ATTR_SIZE (by
557 * llu_setattr_raw()) when file has no bodies. Check this.
559 LASSERT(ergo(ia_valid & ATTR_SIZE, !llu_i2info(inode)->lli_has_smd));
561 if (ia_valid & ATTR_SIZE)
562 st->st_size = attr->ia_size;
563 if (ia_valid & ATTR_UID)
564 st->st_uid = attr->ia_uid;
565 if (ia_valid & ATTR_GID)
566 st->st_gid = attr->ia_gid;
567 if (ia_valid & ATTR_ATIME)
568 st->st_atime = attr->ia_atime;
569 if (ia_valid & ATTR_MTIME)
570 st->st_mtime = attr->ia_mtime;
571 if (ia_valid & ATTR_CTIME)
572 st->st_ctime = attr->ia_ctime;
573 if (ia_valid & ATTR_MODE) {
574 st->st_mode = attr->ia_mode;
575 if (!cfs_curproc_is_in_groups(st->st_gid) &&
576 !cfs_capable(CFS_CAP_FSETID))
577 st->st_mode &= ~S_ISGID;
579 /* mark_inode_dirty(inode); */
583 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
584 struct md_open_data **mod)
587 struct llu_sb_info *sbi = llu_i2sbi(inode);
588 struct ptlrpc_request *request = NULL;
592 llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
593 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
597 ptlrpc_req_finished(request);
598 if (rc != -EPERM && rc != -EACCES)
599 CERROR("md_setattr fails: rc = %d\n", rc);
603 rc = md_get_lustre_md(sbi->ll_md_exp, request,
604 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
606 ptlrpc_req_finished(request);
610 /* We call inode_setattr to adjust timestamps.
611 * If there is at least some data in file, we cleared ATTR_SIZE
612 * above to avoid invoking vmtruncate, otherwise it is important
613 * to call vmtruncate in inode_setattr to update inode->i_size
615 inode_setattr(inode, &op_data->op_attr);
616 llu_update_inode(inode, &md);
617 ptlrpc_req_finished(request);
622 /* Close IO epoch and send Size-on-MDS attribute update. */
623 static int llu_setattr_done_writing(struct inode *inode,
624 struct md_op_data *op_data,
625 struct md_open_data *mod)
627 struct llu_inode_info *lli = llu_i2info(inode);
628 struct intnl_stat *st = llu_i2stat(inode);
632 LASSERT(op_data != NULL);
633 if (!S_ISREG(st->st_mode))
636 /* XXX: pass och here for the recovery purpose. */
637 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
638 op_data->op_ioepoch, PFID(&lli->lli_fid));
640 op_data->op_flags = MF_EPOCH_CLOSE;
641 llu_done_writing_attr(inode, op_data);
642 llu_pack_inode2opdata(inode, op_data, NULL);
644 rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
646 /* MDS has instructed us to obtain Size-on-MDS attribute
647 * from OSTs and send setattr to back to MDS. */
648 rc = llu_som_update(inode, op_data);
650 CERROR("inode %llu mdc truncate failed: rc = %d\n",
651 (unsigned long long)st->st_ino, rc);
656 /* If this inode has objects allocated to it (lsm != NULL), then the OST
657 * object(s) determine the file size and mtime. Otherwise, the MDS will
658 * keep these values until such a time that objects are allocated for it.
659 * We do the MDS operations first, as it is checking permissions for us.
660 * We don't to the MDS RPC if there is nothing that we want to store there,
661 * otherwise there is no harm in updating mtime/atime on the MDS if we are
662 * going to do an RPC anyways.
664 * If we are doing a truncate, we will send the mtime and ctime updates
665 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
666 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
669 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
671 int has_lsm = llu_i2info(inode)->lli_has_smd;
672 struct intnl_stat *st = llu_i2stat(inode);
673 int ia_valid = attr->ia_valid;
674 struct md_op_data op_data = { { 0 } };
675 struct md_open_data *mod = NULL;
679 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
681 if (ia_valid & ATTR_SIZE) {
682 if (attr->ia_size > ll_file_maxbytes(inode)) {
683 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
684 (long long)attr->ia_size,
685 ll_file_maxbytes(inode));
689 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
692 /* We mark all of the fields "set" so MDS/OST does not re-set them */
693 if (attr->ia_valid & ATTR_CTIME) {
694 attr->ia_ctime = CFS_CURRENT_TIME;
695 attr->ia_valid |= ATTR_CTIME_SET;
697 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
698 attr->ia_atime = CFS_CURRENT_TIME;
699 attr->ia_valid |= ATTR_ATIME_SET;
701 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
702 attr->ia_mtime = CFS_CURRENT_TIME;
703 attr->ia_valid |= ATTR_MTIME_SET;
706 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
707 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
708 ", now = "CFS_TIME_T"\n",
709 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
710 LTIME_S(CFS_CURRENT_TIME));
712 /* NB: ATTR_SIZE will only be set after this point if the size
713 * resides on the MDS, ie, this file has no objects. */
715 attr->ia_valid &= ~ATTR_SIZE;
717 /* If only OST attributes being set on objects, don't do MDS RPC.
718 * In that case, we need to check permissions and update the local
719 * inode ourselves so we can call obdo_from_inode() always. */
720 if (ia_valid & (has_lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
721 memcpy(&op_data.op_attr, attr, sizeof(*attr));
723 /* Open epoch for truncate. */
724 if (exp_connect_som(llu_i2mdexp(inode)) &&
725 (ia_valid & ATTR_SIZE))
726 op_data.op_flags = MF_EPOCH_OPEN;
727 rc = llu_md_setattr(inode, &op_data, &mod);
731 llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch);
732 if (!has_lsm || !S_ISREG(st->st_mode)) {
733 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
737 /* The OST doesn't check permissions, but the alternative is
738 * a gratuitous RPC to the MDS. We already rely on the client
739 * to do read/write/truncate permission checks, so is mtime OK?
741 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
742 /* from sys_utime() */
743 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
744 if (current->fsuid != st->st_uid &&
745 (rc = ll_permission(inode, MAY_WRITE)) != 0)
748 /* from inode_change_ok() */
749 if (current->fsuid != st->st_uid &&
750 !cfs_capable(CFS_CAP_FOWNER))
756 /* Won't invoke llu_vmtruncate(), as we already cleared
758 inode_setattr(inode, attr);
761 if (ia_valid & ATTR_SIZE)
762 attr->ia_valid |= ATTR_SIZE;
763 if (ia_valid & (ATTR_SIZE |
764 ATTR_ATIME | ATTR_ATIME_SET |
765 ATTR_MTIME | ATTR_MTIME_SET))
766 /* on truncate and utimes send attributes to osts, setting
767 * mtime/atime to past will be performed under PW 0:EOF extent
768 * lock (new_size:EOF for truncate)
769 * it may seem excessive to send mtime/atime updates to osts
770 * when not setting times to past, but it is necessary due to
771 * possible time de-synchronization */
772 rc = cl_setattr_ost(inode, attr, NULL);
775 if (op_data.op_ioepoch)
776 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
777 return rc ? rc : rc1;
780 /* here we simply act as a thin layer to glue it with
781 * llu_setattr_raw(), which is copy from kernel
783 static int llu_iop_setattr(struct pnode *pno,
786 struct intnl_stat *stbuf)
792 liblustre_wait_event(0);
794 LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
795 SETATTR_UID | SETATTR_GID |
796 SETATTR_LEN | SETATTR_MODE)));
797 memset(&iattr, 0, sizeof(iattr));
799 if (mask & SETATTR_MODE) {
800 iattr.ia_mode = stbuf->st_mode;
801 iattr.ia_valid |= ATTR_MODE;
803 if (mask & SETATTR_MTIME) {
804 iattr.ia_mtime = stbuf->st_mtime;
805 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
807 if (mask & SETATTR_ATIME) {
808 iattr.ia_atime = stbuf->st_atime;
809 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
811 if (mask & SETATTR_UID) {
812 iattr.ia_uid = stbuf->st_uid;
813 iattr.ia_valid |= ATTR_UID;
815 if (mask & SETATTR_GID) {
816 iattr.ia_gid = stbuf->st_gid;
817 iattr.ia_valid |= ATTR_GID;
819 if (mask & SETATTR_LEN) {
820 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
821 iattr.ia_valid |= ATTR_SIZE;
824 iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
825 iattr.ia_ctime = CFS_CURRENT_TIME;
827 rc = llu_setattr_raw(ino, &iattr);
828 liblustre_wait_idle();
832 #define EXT2_LINK_MAX 32000
834 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
836 struct inode *dir = pno->p_base->pb_parent->pb_ino;
837 struct qstr *qstr = &pno->p_base->pb_name;
838 const char *name = qstr->name;
840 struct ptlrpc_request *request = NULL;
841 struct llu_sb_info *sbi = llu_i2sbi(dir);
842 struct md_op_data op_data = {{ 0 }};
846 liblustre_wait_event(0);
847 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
850 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
853 err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
854 S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
855 cfs_curproc_cap_pack(), 0, &request);
856 ptlrpc_req_finished(request);
857 liblustre_wait_event(0);
861 static int llu_readlink_internal(struct inode *inode,
862 struct ptlrpc_request **request,
865 struct llu_inode_info *lli = llu_i2info(inode);
866 struct llu_sb_info *sbi = llu_i2sbi(inode);
867 struct mdt_body *body;
868 struct intnl_stat *st = llu_i2stat(inode);
869 struct md_op_data op_data = {{ 0 }};
870 int rc, symlen = st->st_size + 1;
876 if (lli->lli_symlink_name) {
877 *symname = lli->lli_symlink_name;
878 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
882 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen,
884 op_data.op_valid = OBD_MD_LINKNAME;
886 rc = md_getattr(sbi->ll_md_exp, &op_data, request);
888 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
892 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
893 LASSERT(body != NULL);
895 if ((body->valid & OBD_MD_LINKNAME) == 0) {
896 CERROR ("OBD_MD_LINKNAME not set on reply\n");
897 GOTO (failed, rc = -EPROTO);
900 LASSERT(symlen != 0);
901 if (body->eadatasize != symlen) {
902 CERROR("inode %llu: symlink length %d not expected %d\n",
903 (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
904 GOTO(failed, rc = -EPROTO);
907 *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
908 if (*symname == NULL ||
909 strnlen(*symname, symlen) != symlen - 1) {
910 /* not full/NULL terminated */
911 CERROR("inode %llu: symlink not NULL terminated string"
912 "of length %d\n", (long long)st->st_ino, symlen - 1);
913 GOTO(failed, rc = -EPROTO);
916 OBD_ALLOC(lli->lli_symlink_name, symlen);
917 /* do not return an error if we cannot cache the symlink locally */
918 if (lli->lli_symlink_name)
919 memcpy(lli->lli_symlink_name, *symname, symlen);
924 ptlrpc_req_finished (*request);
928 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
930 struct inode *inode = pno->p_base->pb_ino;
931 struct ptlrpc_request *request;
936 liblustre_wait_event(0);
937 rc = llu_readlink_internal(inode, &request, &symname);
942 strncpy(data, symname, bufsize);
943 rc = strlen(symname);
945 ptlrpc_req_finished(request);
947 liblustre_wait_event(0);
951 static int llu_iop_mknod_raw(struct pnode *pno,
955 struct ptlrpc_request *request = NULL;
956 struct inode *dir = pno->p_parent->p_base->pb_ino;
957 struct llu_sb_info *sbi = llu_i2sbi(dir);
958 struct md_op_data op_data = {{ 0 }};
962 liblustre_wait_event(0);
963 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
964 (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
965 (long long)llu_i2stat(dir)->st_ino);
967 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
970 switch (mode & S_IFMT) {
973 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
978 llu_prep_md_op_data(&op_data, dir, NULL,
979 pno->p_base->pb_name.name,
980 pno->p_base->pb_name.len, 0,
983 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
984 current->fsuid, current->fsgid,
985 cfs_curproc_cap_pack(), dev, &request);
986 ptlrpc_req_finished(request);
994 liblustre_wait_event(0);
998 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
1000 struct inode *src = old->p_base->pb_ino;
1001 struct inode *dir = new->p_parent->p_base->pb_ino;
1002 const char *name = new->p_base->pb_name.name;
1003 int namelen = new->p_base->pb_name.len;
1004 struct ptlrpc_request *request = NULL;
1005 struct md_op_data op_data = {{ 0 }};
1012 liblustre_wait_event(0);
1013 llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1015 rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1016 ptlrpc_req_finished(request);
1017 liblustre_wait_event(0);
1023 * libsysio will clear the inode immediately after return
1025 static int llu_iop_unlink_raw(struct pnode *pno)
1027 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1028 struct qstr *qstr = &pno->p_base->pb_name;
1029 const char *name = qstr->name;
1030 int len = qstr->len;
1031 struct inode *target = pno->p_base->pb_ino;
1032 struct ptlrpc_request *request = NULL;
1033 struct md_op_data op_data = { { 0 } };
1039 liblustre_wait_event(0);
1040 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1042 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1044 rc = llu_objects_destroy(request, dir);
1045 ptlrpc_req_finished(request);
1046 liblustre_wait_idle();
1051 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1053 struct inode *src = old->p_parent->p_base->pb_ino;
1054 struct inode *tgt = new->p_parent->p_base->pb_ino;
1055 const char *oldname = old->p_base->pb_name.name;
1056 int oldnamelen = old->p_base->pb_name.len;
1057 const char *newname = new->p_base->pb_name.name;
1058 int newnamelen = new->p_base->pb_name.len;
1059 struct ptlrpc_request *request = NULL;
1060 struct md_op_data op_data = { { 0 } };
1067 liblustre_wait_event(0);
1068 llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1070 rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1071 oldname, oldnamelen, newname, newnamelen,
1074 rc = llu_objects_destroy(request, src);
1077 ptlrpc_req_finished(request);
1078 liblustre_wait_idle();
1083 #ifdef _HAVE_STATVFS
1084 static int llu_statfs_internal(struct llu_sb_info *sbi,
1085 struct obd_statfs *osfs, __u64 max_age)
1087 struct obd_statfs obd_osfs;
1091 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, 0);
1093 CERROR("md_statfs fails: rc = %d\n", rc);
1097 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1098 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1100 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1101 &obd_statfs, max_age, 0);
1103 CERROR("obd_statfs fails: rc = %d\n", rc);
1107 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1108 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1111 osfs->os_blocks = obd_osfs.os_blocks;
1112 osfs->os_bfree = obd_osfs.os_bfree;
1113 osfs->os_bavail = obd_osfs.os_bavail;
1115 /* If we don't have as many objects free on the OST as inodes
1116 * on the MDS, we reduce the total number of inodes to
1117 * compensate, so that the "inodes in use" number is correct.
1119 if (obd_osfs.os_ffree < osfs->os_ffree) {
1120 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1122 osfs->os_ffree = obd_osfs.os_ffree;
1128 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1130 struct obd_statfs osfs;
1133 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1135 /* For now we will always get up-to-date statfs values, but in the
1136 * future we may allow some amount of caching on the client (e.g.
1137 * from QOS or lprocfs updates). */
1138 rc = llu_statfs_internal(sbi, &osfs,
1139 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS));
1143 statfs_unpack(sfs, &osfs);
1145 if (sizeof(sfs->f_blocks) == 4) {
1146 while (osfs.os_blocks > ~0UL) {
1149 osfs.os_blocks >>= 1;
1150 osfs.os_bfree >>= 1;
1151 osfs.os_bavail >>= 1;
1155 sfs->f_blocks = osfs.os_blocks;
1156 sfs->f_bfree = osfs.os_bfree;
1157 sfs->f_bavail = osfs.os_bavail;
1162 static int llu_iop_statvfs(struct pnode *pno,
1164 struct intnl_statvfs *buf)
1170 liblustre_wait_event(0);
1173 LASSERT(pno->p_base->pb_ino);
1174 rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1178 /* from native driver */
1179 buf->f_bsize = fs.f_bsize; /* file system block size */
1180 buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1181 buf->f_blocks = fs.f_blocks;
1182 buf->f_bfree = fs.f_bfree;
1183 buf->f_bavail = fs.f_bavail;
1184 buf->f_files = fs.f_files; /* Total number serial numbers */
1185 buf->f_ffree = fs.f_ffree; /* Number free serial numbers */
1186 buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1187 buf->f_fsid = fs.f_fsid.__val[1];
1188 buf->f_flag = 0; /* No equiv in statfs; maybe use type? */
1189 buf->f_namemax = fs.f_namelen;
1192 liblustre_wait_event(0);
1195 #endif /* _HAVE_STATVFS */
1197 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1199 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1200 struct qstr *qstr = &pno->p_base->pb_name;
1201 const char *name = qstr->name;
1202 int len = qstr->len;
1203 struct ptlrpc_request *request = NULL;
1204 struct intnl_stat *st = llu_i2stat(dir);
1205 struct md_op_data op_data = {{ 0 }};
1209 liblustre_wait_event(0);
1210 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1211 (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1213 if (st->st_nlink >= EXT2_LINK_MAX)
1216 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1219 err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1220 mode | S_IFDIR, current->fsuid, current->fsgid,
1221 cfs_curproc_cap_pack(), 0, &request);
1222 ptlrpc_req_finished(request);
1223 liblustre_wait_event(0);
1227 static int llu_iop_rmdir_raw(struct pnode *pno)
1229 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1230 struct qstr *qstr = &pno->p_base->pb_name;
1231 const char *name = qstr->name;
1232 int len = qstr->len;
1233 struct ptlrpc_request *request = NULL;
1234 struct md_op_data op_data = {{ 0 }};
1238 liblustre_wait_event(0);
1239 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1240 (long long)llu_i2stat(dir)->st_ino,
1241 llu_i2info(dir)->lli_st_generation, dir);
1243 llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1245 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1246 ptlrpc_req_finished(request);
1248 liblustre_wait_event(0);
1253 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1255 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1257 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1259 /* refer to ll_file_flock() for details */
1260 static int llu_file_flock(struct inode *ino,
1262 struct file_lock *file_lock)
1264 struct llu_inode_info *lli = llu_i2info(ino);
1265 struct intnl_stat *st = llu_i2stat(ino);
1266 struct ldlm_res_id res_id =
1267 { .name = {fid_seq(&lli->lli_fid),
1268 fid_oid(&lli->lli_fid),
1269 fid_ver(&lli->lli_fid),
1271 struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
1272 ldlm_flock_completion_ast, NULL, NULL, file_lock };
1274 struct lustre_handle lockh = {0};
1275 ldlm_policy_data_t flock;
1279 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1280 (unsigned long long)st->st_ino, file_lock);
1282 flock.l_flock.pid = file_lock->fl_pid;
1283 flock.l_flock.start = file_lock->fl_start;
1284 flock.l_flock.end = file_lock->fl_end;
1286 switch (file_lock->fl_type) {
1288 einfo.ei_mode = LCK_PR;
1291 einfo.ei_mode = LCK_NL;
1294 einfo.ei_mode = LCK_PW;
1297 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1304 #if F_SETLKW64 != F_SETLKW
1312 #if F_SETLK64 != F_SETLK
1316 flags = LDLM_FL_BLOCK_NOWAIT;
1320 #if F_GETLK64 != F_GETLK
1324 flags = LDLM_FL_TEST_LOCK;
1325 file_lock->fl_type = einfo.ei_mode;
1328 CERROR("unknown fcntl cmd: %d\n", cmd);
1332 CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#llx, mode=%u, "
1333 "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1334 flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1338 struct lmv_obd *lmv;
1339 struct obd_device *lmv_obd;
1340 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1341 lmv = &lmv_obd->u.lmv;
1343 if (lmv->desc.ld_tgt_count < 1)
1344 RETURN(rc = -ENODEV);
1346 if (lmv->tgts[0] && lmv->tgts[0]->ltd_exp != NULL)
1347 rc = ldlm_cli_enqueue(lmv->tgts[0]->ltd_exp, NULL,
1348 &einfo, &res_id, &flock, &flags,
1349 NULL, 0, LVB_T_NONE, &lockh, 0);
1356 static int assign_type(struct file_lock *fl, int type)
1369 static int flock_to_posix_lock(struct inode *ino,
1370 struct file_lock *fl,
1373 switch (l->l_whence) {
1374 /* XXX: only SEEK_SET is supported in lustre */
1382 fl->fl_end = l->l_len - 1;
1386 fl->fl_end = OFFSET_MAX;
1388 fl->fl_pid = getpid();
1389 fl->fl_flags = FL_POSIX;
1390 fl->fl_notify = NULL;
1391 fl->fl_insert = NULL;
1392 fl->fl_remove = NULL;
1393 /* XXX: these fields can't be filled with suitable values,
1394 but I think lustre doesn't use them.
1396 fl->fl_owner = NULL;
1399 return assign_type(fl, l->l_type);
1402 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1404 struct file_lock fl;
1408 if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1411 error = flock_to_posix_lock(ino, &fl, flock);
1415 error = llu_file_flock(ino, F_GETLK, &fl);
1419 flock->l_type = F_UNLCK;
1420 if (fl.fl_type != F_UNLCK) {
1421 flock->l_pid = fl.fl_pid;
1422 flock->l_start = fl.fl_start;
1423 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1424 fl.fl_end - fl.fl_start + 1;
1425 flock->l_whence = SEEK_SET;
1426 flock->l_type = fl.fl_type;
1433 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1435 struct file_lock fl;
1436 int flags = llu_i2info(ino)->lli_open_flags + 1;
1439 error = flock_to_posix_lock(ino, &fl, flock);
1442 if (cmd == F_SETLKW)
1443 fl.fl_flags |= FL_SLEEP;
1446 switch (flock->l_type) {
1448 if (!(flags & FMODE_READ))
1452 if (!(flags & FMODE_WRITE))
1462 error = llu_file_flock(ino, cmd, &fl);
1470 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1472 struct llu_inode_info *lli = llu_i2info(ino);
1474 struct flock *flock;
1477 liblustre_wait_event(0);
1480 *rtn = lli->lli_open_flags;
1483 flags = va_arg(ap, long);
1484 flags &= FCNTL_FLMASK;
1485 if (flags & FCNTL_FLMASK_INVALID) {
1486 LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1487 "the O_NONBLOCK or O_ASYNC flags. "
1488 "Please fix your application.\n");
1493 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1494 (lli->lli_open_flags & ~FCNTL_FLMASK);
1499 #if F_GETLK64 != F_GETLK
1503 flock = va_arg(ap, struct flock *);
1504 err = llu_fcntl_getlk(ino, flock);
1509 #if F_SETLKW64 != F_SETLKW
1515 #if F_SETLK64 != F_SETLK
1519 flock = va_arg(ap, struct flock *);
1520 err = llu_fcntl_setlk(ino, cmd, flock);
1524 CERROR("unsupported fcntl cmd %x\n", cmd);
1530 liblustre_wait_event(0);
1534 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1536 struct llu_inode_info *lli = llu_i2info(inode);
1537 struct ll_file_data *fd = lli->lli_file_data;
1539 struct ccc_grouplock grouplock;
1542 if (fd->fd_flags & LL_FILE_IGNORE_LOCK) {
1545 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1548 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1550 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1551 arg, (lli->lli_open_flags & O_NONBLOCK),
1557 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1558 fd->fd_grouplock = grouplock;
1563 int llu_put_grouplock(struct inode *inode, unsigned long arg)
1565 struct llu_inode_info *lli = llu_i2info(inode);
1566 struct ll_file_data *fd = lli->lli_file_data;
1567 struct ccc_grouplock grouplock;
1570 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1573 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1575 if (fd->fd_grouplock.cg_gid != arg)
1578 grouplock = fd->fd_grouplock;
1579 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1580 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1582 cl_put_grouplock(&grouplock);
1587 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1589 struct llu_sb_info *sbi = llu_i2sbi(ino);
1590 struct ptlrpc_request *request = NULL;
1591 struct md_op_data op_data = {{ 0 }};
1592 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1595 llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1598 LASSERT(sizeof(lum) == sizeof(*lump));
1599 LASSERT(sizeof(lum.lmm_objects[0]) ==
1600 sizeof(lump->lmm_objects[0]));
1601 if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1604 switch (lum.lmm_magic) {
1605 case LOV_USER_MAGIC_V1: {
1606 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1607 lustre_swab_lov_user_md_v1(&lum);
1610 case LOV_USER_MAGIC_V3: {
1611 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1612 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1616 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1617 " %#08x != %#08x nor %#08x\n",
1618 lum.lmm_magic, LOV_USER_MAGIC_V1,
1624 /* swabbing is done in lov_setstripe() on server side */
1625 rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1626 sizeof(lum), NULL, 0, &request, NULL);
1628 ptlrpc_req_finished(request);
1629 if (rc != -EPERM && rc != -EACCES)
1630 CERROR("md_setattr fails: rc = %d\n", rc);
1633 ptlrpc_req_finished(request);
1638 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1639 struct lov_user_md *lum, int lum_size)
1641 struct llu_sb_info *sbi = llu_i2sbi(ino);
1642 struct llu_inode_info *lli = llu_i2info(ino);
1643 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1644 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR,
1645 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
1646 struct ptlrpc_request *req = NULL;
1647 struct lustre_md md;
1648 struct md_op_data data = {{ 0 }};
1649 struct lustre_handle lockh;
1653 if (lli->lli_has_smd) {
1654 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1655 PFID(&lli->lli_fid));
1659 llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1661 rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1662 &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1666 req = oit.d.lustre.it_data;
1667 rc = it_open_error(DISP_IT_EXECD, &oit);
1673 rc = it_open_error(DISP_OPEN_OPEN, &oit);
1679 rc = md_get_lustre_md(sbi->ll_md_exp, req,
1680 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1684 llu_update_inode(ino, &md);
1685 llu_local_open(lli, &oit);
1686 /* release intent */
1687 if (lustre_handle_is_used(&lockh))
1688 ldlm_lock_decref(&lockh, LCK_CR);
1689 ptlrpc_req_finished(req);
1691 rc = llu_file_release(ino);
1696 ptlrpc_req_finished(req);
1700 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1702 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1704 int flags = FMODE_WRITE;
1707 LASSERT(sizeof(lum) == sizeof(*lump));
1708 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1709 if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1712 rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1716 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1718 struct intnl_stat *st = llu_i2stat(ino);
1719 if (S_ISREG(st->st_mode))
1720 return llu_lov_file_setstripe(ino, arg);
1721 if (S_ISDIR(st->st_mode))
1722 return llu_lov_dir_setstripe(ino, arg);
1727 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1729 struct lov_stripe_md *lsm = NULL;
1732 lsm = ccc_inode_lsm_get(ino);
1734 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1736 ccc_inode_lsm_put(ino, lsm);
1740 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1746 liblustre_wait_event(0);
1749 case LL_IOC_GROUP_LOCK:
1750 arg = va_arg(ap, unsigned long);
1751 rc = llu_get_grouplock(ino, arg);
1753 case LL_IOC_GROUP_UNLOCK:
1754 arg = va_arg(ap, unsigned long);
1755 rc = llu_put_grouplock(ino, arg);
1757 case LL_IOC_LOV_SETSTRIPE:
1758 arg = va_arg(ap, unsigned long);
1759 rc = llu_lov_setstripe(ino, arg);
1761 case LL_IOC_LOV_GETSTRIPE:
1762 arg = va_arg(ap, unsigned long);
1763 rc = llu_lov_getstripe(ino, arg);
1766 CERROR("did not support ioctl cmd %lx\n", request);
1771 liblustre_wait_event(0);
1776 * we already do syncronous read/write
1778 static int llu_iop_sync(struct inode *inode)
1780 liblustre_wait_event(0);
1784 static int llu_iop_datasync(struct inode *inode)
1786 liblustre_wait_event(0);
1790 struct filesys_ops llu_filesys_ops =
1792 fsop_gone: llu_fsop_gone,
1795 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1797 struct inode *inode;
1799 struct file_identifier fileid = {&fid, sizeof(fid)};
1801 if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1802 (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1803 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1805 return ERR_PTR(-EPERM);
1808 /* try to find existing inode */
1809 fid = md->body->fid1;
1811 inode = _sysio_i_find(fs, &fileid);
1813 if (inode->i_zombie/* ||
1814 lli->lli_st_generation != md->body->generation*/) {
1818 llu_update_inode(inode, md);
1823 inode = llu_new_inode(fs, &fid);
1825 llu_update_inode(inode, md);
1831 llu_fsswop_mount(const char *source,
1833 const void *data __IS_UNUSED,
1834 struct pnode *tocover,
1835 struct mount **mntp)
1839 struct pnode_base *rootpb;
1840 struct obd_device *obd;
1841 struct llu_sb_info *sbi;
1842 struct obd_statfs osfs;
1843 static struct qstr noname = { NULL, 0, 0 };
1844 struct ptlrpc_request *request = NULL;
1845 struct lustre_md md;
1847 struct config_llog_instance cfg = {0, };
1848 struct lustre_profile *lprof;
1849 char *zconf_mgsnid, *zconf_profile;
1850 char *osc = NULL, *mdc = NULL;
1851 int async = 1, err = -EINVAL;
1852 struct obd_connect_data ocd = {0,};
1853 struct md_op_data op_data = {{0}};
1854 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1855 const int instlen = sizeof(cfg.cfg_instance) * 2 + 2;
1859 if (ll_parse_mount_target(source,
1862 CERROR("mal-formed target %s\n", source);
1865 if (!zconf_mgsnid || !zconf_profile) {
1866 printf("Liblustre: invalid target %s\n", source);
1869 /* allocate & initialize sbi */
1870 OBD_ALLOC(sbi, sizeof(*sbi));
1874 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
1875 ll_generate_random_uuid(uuid);
1876 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1878 /* generate a string unique to this super, let's try
1879 the address of the super itself.*/
1880 cfg.cfg_instance = sbi;
1882 /* retrive & parse config log */
1883 cfg.cfg_uuid = sbi->ll_sb_uuid;
1884 err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
1886 CERROR("Unable to process log: %s\n", zconf_profile);
1887 GOTO(out_free, err);
1890 lprof = class_get_profile(zconf_profile);
1891 if (lprof == NULL) {
1892 CERROR("No profile found: %s\n", zconf_profile);
1893 GOTO(out_free, err = -EINVAL);
1895 OBD_ALLOC(osc, strlen(lprof->lp_dt) + instlen + 2);
1896 sprintf(osc, "%s-%p", lprof->lp_dt, cfg.cfg_instance);
1898 OBD_ALLOC(mdc, strlen(lprof->lp_md) + instlen + 2);
1899 sprintf(mdc, "%s-%p", lprof->lp_md, cfg.cfg_instance);
1903 GOTO(out_free, err = -EINVAL);
1907 GOTO(out_free, err = -EINVAL);
1910 fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1916 obd = class_name2obd(mdc);
1918 CERROR("MDC %s: not setup or attached\n", mdc);
1919 GOTO(out_free, err = -EINVAL);
1921 obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC),
1922 KEY_ASYNC, sizeof(async), &async, NULL);
1924 ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
1925 OBD_CONNECT_FID | OBD_CONNECT_AT |
1926 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
1927 OBD_CONNECT_LVB_TYPE;
1929 #ifdef LIBLUSTRE_POSIX_ACL
1930 ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
1932 ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
1933 ocd.ocd_version = LUSTRE_VERSION_CODE;
1936 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1938 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
1939 GOTO(out_free, err);
1942 err = obd_statfs(NULL, sbi->ll_md_exp, &osfs, 100000000, 0);
1947 * FIXME fill fs stat data into sbi here!!! FIXME
1951 obd = class_name2obd(osc);
1953 CERROR("OSC %s: not setup or attached\n", osc);
1954 GOTO(out_md, err = -EINVAL);
1956 obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC),
1957 KEY_ASYNC, sizeof(async), &async, NULL);
1959 obd->obd_upcall.onu_owner = &sbi->ll_lco;
1960 obd->obd_upcall.onu_upcall = cl_ocd_update;
1962 ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
1963 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
1964 OBD_CONNECT_FID | OBD_CONNECT_AT |
1965 OBD_CONNECT_FULL20 | OBD_CONNECT_EINPROGRESS |
1966 OBD_CONNECT_LVB_TYPE;
1968 ocd.ocd_version = LUSTRE_VERSION_CODE;
1969 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1971 CERROR("cannot connect to %s: rc = %d\n", osc, err);
1974 sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
1975 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
1976 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
1978 fid_zero(&sbi->ll_root_fid);
1979 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
1981 CERROR("cannot mds_connect: rc = %d\n", err);
1982 GOTO(out_lock_cn_cb, err);
1984 if (!fid_is_sane(&sbi->ll_root_fid)) {
1985 CERROR("Invalid root fid during mount\n");
1986 GOTO(out_lock_cn_cb, err = -EINVAL);
1988 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
1990 op_data.op_fid1 = sbi->ll_root_fid;
1991 op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
1992 /* fetch attr of root inode */
1993 err = md_getattr(sbi->ll_md_exp, &op_data, &request);
1995 CERROR("md_getattr failed for root: rc = %d\n", err);
1996 GOTO(out_lock_cn_cb, err);
1999 err = md_get_lustre_md(sbi->ll_md_exp, request,
2000 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2002 CERROR("failed to understand root inode md: rc = %d\n",err);
2003 GOTO(out_request, err);
2006 LASSERT(fid_is_sane(&sbi->ll_root_fid));
2008 root = llu_iget(fs, &md);
2009 if (!root || IS_ERR(root)) {
2010 CERROR("fail to generate root inode\n");
2011 GOTO(out_request, err = -EBADF);
2015 * Generate base path-node for root.
2017 rootpb = _sysio_pb_new(&noname, NULL, root);
2023 err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2025 _sysio_pb_gone(rootpb);
2031 ptlrpc_req_finished(request);
2033 CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2038 _sysio_i_gone(root);
2040 ptlrpc_req_finished(request);
2042 obd_disconnect(sbi->ll_dt_exp);
2044 obd_disconnect(sbi->ll_md_exp);
2047 OBD_FREE(osc, strlen(lprof->lp_dt) + instlen + 2);
2049 OBD_FREE(mdc, strlen(lprof->lp_md) + instlen + 2);
2050 OBD_FREE(sbi, sizeof(*sbi));
2051 liblustre_wait_idle();
2055 struct fssw_ops llu_fssw_ops = {
2059 static struct inode_ops llu_inode_ops = {
2060 inop_lookup: llu_iop_lookup,
2061 inop_getattr: llu_iop_getattr,
2062 inop_setattr: llu_iop_setattr,
2063 inop_filldirentries: llu_iop_filldirentries,
2064 inop_mkdir: llu_iop_mkdir_raw,
2065 inop_rmdir: llu_iop_rmdir_raw,
2066 inop_symlink: llu_iop_symlink_raw,
2067 inop_readlink: llu_iop_readlink,
2068 inop_open: llu_iop_open,
2069 inop_close: llu_iop_close,
2070 inop_link: llu_iop_link_raw,
2071 inop_unlink: llu_iop_unlink_raw,
2072 inop_rename: llu_iop_rename_raw,
2073 inop_pos: llu_iop_pos,
2074 inop_read: llu_iop_read,
2075 inop_write: llu_iop_write,
2076 inop_iodone: llu_iop_iodone,
2077 inop_fcntl: llu_iop_fcntl,
2078 inop_sync: llu_iop_sync,
2079 inop_datasync: llu_iop_datasync,
2080 inop_ioctl: llu_iop_ioctl,
2081 inop_mknod: llu_iop_mknod_raw,
2082 #ifdef _HAVE_STATVFS
2083 inop_statvfs: llu_iop_statvfs,
2085 inop_gone: llu_iop_gone,