1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/liblustre/super.c
40 * Lustre Light Super operations
43 #define DEBUG_SUBSYSTEM S_LLITE
49 #include <sys/types.h>
52 #include <sys/queue.h>
54 # include <sys/statvfs.h>
56 # include <sys/statfs.h>
59 #include "llite_lib.h"
67 #define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH)
69 static int ll_permission(struct inode *inode, int mask)
71 struct intnl_stat *st = llu_i2stat(inode);
72 mode_t mode = st->st_mode;
74 if (current->fsuid == st->st_uid)
76 else if (cfs_curproc_is_in_groups(st->st_gid))
79 if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
82 if ((mask & (MAY_READ|MAY_WRITE)) ||
83 (st->st_mode & S_IXUGO))
84 if (cfs_capable(CFS_CAP_DAC_OVERRIDE))
87 if (mask == MAY_READ ||
88 (S_ISDIR(st->st_mode) && !(mask & MAY_WRITE))) {
89 if (cfs_capable(CFS_CAP_DAC_READ_SEARCH))
96 static void llu_fsop_gone(struct filesys *fs)
98 struct llu_sb_info *sbi = (struct llu_sb_info *) fs->fs_private;
99 struct obd_device *obd = class_exp2obd(sbi->ll_md_exp);
103 cfs_list_del(&sbi->ll_conn_chain);
105 obd_disconnect(sbi->ll_dt_exp);
106 obd_disconnect(sbi->ll_md_exp);
108 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
109 class_manual_cleanup(obd);
111 OBD_FREE(sbi, sizeof(*sbi));
113 liblustre_wait_idle();
117 static struct inode_ops llu_inode_ops;
119 static ldlm_mode_t llu_take_md_lock(struct inode *inode, __u64 bits,
120 struct lustre_handle *lockh)
122 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
128 fid = &llu_i2info(inode)->lli_fid;
129 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
131 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
132 rc = md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
133 LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
137 void llu_update_inode(struct inode *inode, struct lustre_md *md)
139 struct llu_inode_info *lli = llu_i2info(inode);
140 struct mdt_body *body = md->body;
141 struct lov_stripe_md *lsm = md->lsm;
142 struct intnl_stat *st = llu_i2stat(inode);
144 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
146 if (body->valid & OBD_MD_FLMODE)
147 st->st_mode = (st->st_mode & S_IFMT)|(body->mode & ~S_IFMT);
148 if (body->valid & OBD_MD_FLTYPE)
149 st->st_mode = (st->st_mode & ~S_IFMT)|(body->mode & S_IFMT);
152 if (lli->lli_smd == NULL) {
153 cl_file_inode_init(inode, md);
155 lli->lli_maxbytes = lsm->lsm_maxbytes;
156 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
157 lli->lli_maxbytes = MAX_LFS_FILESIZE;
159 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
160 CERROR("lsm mismatch for inode %lld\n",
161 (long long)st->st_ino);
167 if (body->valid & OBD_MD_FLATIME) {
168 if (body->atime > LTIME_S(st->st_atime))
169 LTIME_S(st->st_atime) = body->atime;
170 lli->lli_lvb.lvb_atime = body->atime;
172 if (body->valid & OBD_MD_FLMTIME) {
173 if (body->mtime > LTIME_S(st->st_mtime))
174 LTIME_S(st->st_mtime) = body->mtime;
175 lli->lli_lvb.lvb_mtime = body->mtime;
177 if (body->valid & OBD_MD_FLCTIME) {
178 if (body->ctime > LTIME_S(st->st_ctime))
179 LTIME_S(st->st_ctime) = body->ctime;
180 lli->lli_lvb.lvb_ctime = body->ctime;
182 if (S_ISREG(st->st_mode))
183 st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE);
185 st->st_blksize = 4096;
186 if (body->valid & OBD_MD_FLUID)
187 st->st_uid = body->uid;
188 if (body->valid & OBD_MD_FLGID)
189 st->st_gid = body->gid;
190 if (body->valid & OBD_MD_FLNLINK)
191 st->st_nlink = body->nlink;
192 if (body->valid & OBD_MD_FLRDEV)
193 st->st_rdev = body->rdev;
194 if (body->valid & OBD_MD_FLFLAGS)
195 lli->lli_st_flags = body->flags;
196 if (body->valid & OBD_MD_FLSIZE) {
197 if ((llu_i2sbi(inode)->ll_lco.lco_flags & OBD_CONNECT_SOM) &&
198 S_ISREG(st->st_mode) && lli->lli_smd) {
199 struct lustre_handle lockh;
202 /* As it is possible a blocking ast has been processed
203 * by this time, we need to check there is an UPDATE
204 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
206 mode = llu_take_md_lock(inode, MDS_INODELOCK_UPDATE,
209 st->st_size = body->size;
210 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
211 ldlm_lock_decref(&lockh, mode);
214 st->st_size = body->size;
217 if (body->valid & OBD_MD_FLBLOCKS)
218 st->st_blocks = body->blocks;
222 void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid)
224 struct llu_inode_info *lli = llu_i2info(dst);
225 struct intnl_stat *st = llu_i2stat(dst);
227 valid &= src->o_valid;
229 LASSERTF(!(valid & (OBD_MD_FLTYPE | OBD_MD_FLGENER | OBD_MD_FLFID |
230 OBD_MD_FLID | OBD_MD_FLGROUP)),
231 "object "LPU64"/"LPU64", valid %x\n",
232 src->o_id, src->o_seq, valid);
234 if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
235 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
238 LTIME_S(st->st_mtime), LTIME_S(st->st_ctime),
239 (long)src->o_mtime, (long)src->o_ctime);
241 if (valid & OBD_MD_FLATIME)
242 LTIME_S(st->st_atime) = src->o_atime;
243 if (valid & OBD_MD_FLMTIME)
244 LTIME_S(st->st_mtime) = src->o_mtime;
245 if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
246 LTIME_S(st->st_ctime) = src->o_ctime;
247 if (valid & OBD_MD_FLSIZE)
248 st->st_size = src->o_size;
249 if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
250 st->st_blocks = src->o_blocks;
251 if (valid & OBD_MD_FLBLKSZ)
252 st->st_blksize = src->o_blksize;
253 if (valid & OBD_MD_FLTYPE)
254 st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
255 if (valid & OBD_MD_FLMODE)
256 st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
257 if (valid & OBD_MD_FLUID)
258 st->st_uid = src->o_uid;
259 if (valid & OBD_MD_FLGID)
260 st->st_gid = src->o_gid;
261 if (valid & OBD_MD_FLFLAGS)
262 lli->lli_st_flags = src->o_flags;
266 * Performs the getattr on the inode and updates its fields.
267 * If @sync != 0, perform the getattr under the server-side lock.
269 int llu_inode_getattr(struct inode *inode, struct obdo *obdo,
270 __u64 ioepoch, int sync)
272 struct llu_inode_info *lli = llu_i2info(inode);
273 struct ptlrpc_request_set *set;
274 struct lov_stripe_md *lsm = lli->lli_smd;
275 struct obd_info oinfo = { { { 0 } } };
283 oinfo.oi_oa->o_id = lsm->lsm_object_id;
284 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
285 oinfo.oi_oa->o_mode = S_IFREG;
286 oinfo.oi_oa->o_ioepoch = ioepoch;
287 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
288 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
289 OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
290 OBD_MD_FLCTIME | OBD_MD_FLGROUP |
291 OBD_MD_FLATIME | OBD_MD_FLEPOCH;
292 obdo_set_parent_fid(oinfo.oi_oa, &llu_i2info(inode)->lli_fid);
294 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
295 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
298 set = ptlrpc_prep_set();
300 CERROR ("ENOMEM allocing request set\n");
303 rc = obd_getattr_async(llu_i2obdexp(inode), &oinfo, set);
305 rc = ptlrpc_set_wait(set);
306 ptlrpc_set_destroy(set);
311 oinfo.oi_oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
312 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
315 obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
316 CDEBUG(D_INODE, "objid "LPX64" size %llu, blocks %llu, "
317 "blksize %llu\n", lli->lli_smd->lsm_object_id,
318 (long long unsigned)llu_i2stat(inode)->st_size,
319 (long long unsigned)llu_i2stat(inode)->st_blocks,
320 (long long unsigned)llu_i2stat(inode)->st_blksize);
324 static struct inode* llu_new_inode(struct filesys *fs,
328 struct llu_inode_info *lli;
329 struct intnl_stat st = {
332 #ifndef AUTOMOUNT_FILE_NAME
333 .st_mode = fid->f_type & S_IFMT,
335 .st_mode = fid->f_type /* all of the bits! */
338 /* FIXME: fix this later */
345 OBD_ALLOC(lli, sizeof(*lli));
349 /* initialize lli here */
350 lli->lli_sbi = llu_fs2sbi(fs);
352 lli->lli_symlink_name = NULL;
354 lli->lli_maxbytes = (__u64)(~0UL);
355 lli->lli_file_data = NULL;
357 lli->lli_sysio_fid.fid_data = &lli->lli_fid;
358 lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
361 /* file identifier is needed by functions like _sysio_i_find() */
362 inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
363 &st, 0, &llu_inode_ops, lli);
366 OBD_FREE(lli, sizeof(*lli));
371 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
373 struct lustre_handle lockh;
374 ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
381 fid = &llu_i2info(inode)->lli_fid;
382 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
384 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
385 if (md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
386 LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
392 static int llu_inode_revalidate(struct inode *inode)
394 struct llu_inode_info *lli = llu_i2info(inode);
395 struct intnl_stat *st = llu_i2stat(inode);
399 CERROR("REPORT THIS LINE TO PETER\n");
403 if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
405 struct ptlrpc_request *req = NULL;
406 struct llu_sb_info *sbi = llu_i2sbi(inode);
407 struct md_op_data op_data = { { 0 } };
408 unsigned long valid = OBD_MD_FLGETATTR;
411 /* Why don't we update all valid MDS fields here, if we're
412 * doing an RPC anyways? -phil */
413 if (S_ISREG(st->st_mode)) {
414 ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
415 valid |= OBD_MD_FLEASIZE;
418 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen,
420 op_data.op_valid = valid;
422 rc = md_getattr(sbi->ll_md_exp, &op_data, &req);
424 CERROR("failure %d inode %llu\n", rc,
425 (long long)st->st_ino);
428 rc = md_get_lustre_md(sbi->ll_md_exp, req,
429 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
431 /* XXX Too paranoid? */
432 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
433 !((md.body->valid & OBD_MD_FLNLINK) &&
434 (md.body->nlink == 0))) {
435 CERROR("Asked for %s eadata but got %s (%d)\n",
436 (valid & OBD_MD_FLEASIZE) ? "some" : "no",
437 (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
438 md.body->eadatasize);
441 ptlrpc_req_finished(req);
446 llu_update_inode(inode, &md);
447 if (md.lsm != NULL && lli->lli_smd != md.lsm)
448 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
449 ptlrpc_req_finished(req);
453 /* object not yet allocated, don't validate size */
454 st->st_atime = lli->lli_lvb.lvb_atime;
455 st->st_mtime = lli->lli_lvb.lvb_mtime;
456 st->st_ctime = lli->lli_lvb.lvb_ctime;
460 /* ll_glimpse_size will prefer locally cached writes if they extend
462 RETURN(cl_glimpse_size(inode));
465 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
467 *b = *llu_i2stat(ino);
470 static int llu_iop_getattr(struct pnode *pno,
472 struct intnl_stat *b)
477 liblustre_wait_event(0);
481 LASSERT(pno->p_base->pb_ino);
482 ino = pno->p_base->pb_ino;
484 LASSERT(!pno || pno->p_base->pb_ino == ino);
487 /* libsysio might call us directly without intent lock,
488 * we must re-fetch the attrs here
490 rc = llu_inode_revalidate(ino);
492 copy_stat_buf(ino, b);
493 LASSERT(!llu_i2info(ino)->lli_it);
496 liblustre_wait_event(0);
500 static int null_if_equal(struct ldlm_lock *lock, void *data)
502 if (data == lock->l_ast_data) {
503 lock->l_ast_data = NULL;
505 if (lock->l_req_mode != lock->l_granted_mode)
506 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
509 return LDLM_ITER_CONTINUE;
512 void llu_clear_inode(struct inode *inode)
514 struct llu_inode_info *lli = llu_i2info(inode);
515 struct llu_sb_info *sbi = llu_i2sbi(inode);
518 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
519 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
522 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
523 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
524 null_if_equal, inode);
527 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
528 null_if_equal, inode);
530 cl_inode_fini(inode);
533 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
537 if (lli->lli_symlink_name) {
538 OBD_FREE(lli->lli_symlink_name,
539 strlen(lli->lli_symlink_name) + 1);
540 lli->lli_symlink_name = NULL;
546 void llu_iop_gone(struct inode *inode)
548 struct llu_inode_info *lli = llu_i2info(inode);
551 liblustre_wait_event(0);
552 llu_clear_inode(inode);
554 OBD_FREE(lli, sizeof(*lli));
558 static int inode_setattr(struct inode * inode, struct iattr * attr)
560 unsigned int ia_valid = attr->ia_valid;
561 struct intnl_stat *st = llu_i2stat(inode);
565 * inode_setattr() is only ever invoked with ATTR_SIZE (by
566 * llu_setattr_raw()) when file has no bodies. Check this.
568 LASSERT(ergo(ia_valid & ATTR_SIZE, llu_i2info(inode)->lli_smd == NULL));
570 if (ia_valid & ATTR_SIZE)
571 st->st_size = attr->ia_size;
572 if (ia_valid & ATTR_UID)
573 st->st_uid = attr->ia_uid;
574 if (ia_valid & ATTR_GID)
575 st->st_gid = attr->ia_gid;
576 if (ia_valid & ATTR_ATIME)
577 st->st_atime = attr->ia_atime;
578 if (ia_valid & ATTR_MTIME)
579 st->st_mtime = attr->ia_mtime;
580 if (ia_valid & ATTR_CTIME)
581 st->st_ctime = attr->ia_ctime;
582 if (ia_valid & ATTR_MODE) {
583 st->st_mode = attr->ia_mode;
584 if (!cfs_curproc_is_in_groups(st->st_gid) &&
585 !cfs_capable(CFS_CAP_FSETID))
586 st->st_mode &= ~S_ISGID;
588 /* mark_inode_dirty(inode); */
592 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
593 struct md_open_data **mod)
596 struct llu_sb_info *sbi = llu_i2sbi(inode);
597 struct ptlrpc_request *request = NULL;
601 llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
602 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
606 ptlrpc_req_finished(request);
607 if (rc != -EPERM && rc != -EACCES)
608 CERROR("md_setattr fails: rc = %d\n", rc);
612 rc = md_get_lustre_md(sbi->ll_md_exp, request,
613 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
615 ptlrpc_req_finished(request);
619 /* We call inode_setattr to adjust timestamps.
620 * If there is at least some data in file, we cleared ATTR_SIZE
621 * above to avoid invoking vmtruncate, otherwise it is important
622 * to call vmtruncate in inode_setattr to update inode->i_size
624 inode_setattr(inode, &op_data->op_attr);
625 llu_update_inode(inode, &md);
626 ptlrpc_req_finished(request);
631 /* Close IO epoch and send Size-on-MDS attribute update. */
632 static int llu_setattr_done_writing(struct inode *inode,
633 struct md_op_data *op_data,
634 struct md_open_data *mod)
636 struct llu_inode_info *lli = llu_i2info(inode);
637 struct intnl_stat *st = llu_i2stat(inode);
641 LASSERT(op_data != NULL);
642 if (!S_ISREG(st->st_mode))
645 /* XXX: pass och here for the recovery purpose. */
646 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
647 op_data->op_ioepoch, PFID(&lli->lli_fid));
649 op_data->op_flags = MF_EPOCH_CLOSE;
650 llu_done_writing_attr(inode, op_data);
651 llu_pack_inode2opdata(inode, op_data, NULL);
653 rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
655 /* MDS has instructed us to obtain Size-on-MDS attribute
656 * from OSTs and send setattr to back to MDS. */
657 rc = llu_som_update(inode, op_data);
659 CERROR("inode %llu mdc truncate failed: rc = %d\n",
660 (unsigned long long)st->st_ino, rc);
665 /* If this inode has objects allocated to it (lsm != NULL), then the OST
666 * object(s) determine the file size and mtime. Otherwise, the MDS will
667 * keep these values until such a time that objects are allocated for it.
668 * We do the MDS operations first, as it is checking permissions for us.
669 * We don't to the MDS RPC if there is nothing that we want to store there,
670 * otherwise there is no harm in updating mtime/atime on the MDS if we are
671 * going to do an RPC anyways.
673 * If we are doing a truncate, we will send the mtime and ctime updates
674 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
675 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
678 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
680 struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd;
681 struct intnl_stat *st = llu_i2stat(inode);
682 int ia_valid = attr->ia_valid;
683 struct md_op_data op_data = { { 0 } };
684 struct md_open_data *mod = NULL;
688 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
690 if (ia_valid & ATTR_SIZE) {
691 if (attr->ia_size > ll_file_maxbytes(inode)) {
692 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
693 (long long)attr->ia_size,
694 ll_file_maxbytes(inode));
698 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
701 /* We mark all of the fields "set" so MDS/OST does not re-set them */
702 if (attr->ia_valid & ATTR_CTIME) {
703 attr->ia_ctime = CFS_CURRENT_TIME;
704 attr->ia_valid |= ATTR_CTIME_SET;
706 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
707 attr->ia_atime = CFS_CURRENT_TIME;
708 attr->ia_valid |= ATTR_ATIME_SET;
710 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
711 attr->ia_mtime = CFS_CURRENT_TIME;
712 attr->ia_valid |= ATTR_MTIME_SET;
715 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
716 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
717 ", now = "CFS_TIME_T"\n",
718 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
719 LTIME_S(CFS_CURRENT_TIME));
721 /* NB: ATTR_SIZE will only be set after this point if the size
722 * resides on the MDS, ie, this file has no objects. */
724 attr->ia_valid &= ~ATTR_SIZE;
726 /* If only OST attributes being set on objects, don't do MDS RPC.
727 * In that case, we need to check permissions and update the local
728 * inode ourselves so we can call obdo_from_inode() always. */
729 if (ia_valid & (lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
730 memcpy(&op_data.op_attr, attr, sizeof(*attr));
732 /* Open epoch for truncate. */
733 if (exp_connect_som(llu_i2mdexp(inode)) &&
734 (ia_valid & ATTR_SIZE))
735 op_data.op_flags = MF_EPOCH_OPEN;
736 rc = llu_md_setattr(inode, &op_data, &mod);
740 llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch);
741 if (!lsm || !S_ISREG(st->st_mode)) {
742 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
746 /* The OST doesn't check permissions, but the alternative is
747 * a gratuitous RPC to the MDS. We already rely on the client
748 * to do read/write/truncate permission checks, so is mtime OK?
750 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
751 /* from sys_utime() */
752 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
753 if (current->fsuid != st->st_uid &&
754 (rc = ll_permission(inode, MAY_WRITE)) != 0)
757 /* from inode_change_ok() */
758 if (current->fsuid != st->st_uid &&
759 !cfs_capable(CFS_CAP_FOWNER))
765 /* Won't invoke llu_vmtruncate(), as we already cleared
767 inode_setattr(inode, attr);
770 if (ia_valid & ATTR_SIZE)
771 attr->ia_valid |= ATTR_SIZE;
772 if (ia_valid & (ATTR_SIZE |
773 ATTR_ATIME | ATTR_ATIME_SET |
774 ATTR_MTIME | ATTR_MTIME_SET))
775 /* on truncate and utimes send attributes to osts, setting
776 * mtime/atime to past will be performed under PW 0:EOF extent
777 * lock (new_size:EOF for truncate)
778 * it may seem excessive to send mtime/atime updates to osts
779 * when not setting times to past, but it is necessary due to
780 * possible time de-synchronization */
781 rc = cl_setattr_ost(inode, attr, NULL);
784 if (op_data.op_ioepoch)
785 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
786 return rc ? rc : rc1;
789 /* here we simply act as a thin layer to glue it with
790 * llu_setattr_raw(), which is copy from kernel
792 static int llu_iop_setattr(struct pnode *pno,
795 struct intnl_stat *stbuf)
801 liblustre_wait_event(0);
803 LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
804 SETATTR_UID | SETATTR_GID |
805 SETATTR_LEN | SETATTR_MODE)));
806 memset(&iattr, 0, sizeof(iattr));
808 if (mask & SETATTR_MODE) {
809 iattr.ia_mode = stbuf->st_mode;
810 iattr.ia_valid |= ATTR_MODE;
812 if (mask & SETATTR_MTIME) {
813 iattr.ia_mtime = stbuf->st_mtime;
814 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
816 if (mask & SETATTR_ATIME) {
817 iattr.ia_atime = stbuf->st_atime;
818 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
820 if (mask & SETATTR_UID) {
821 iattr.ia_uid = stbuf->st_uid;
822 iattr.ia_valid |= ATTR_UID;
824 if (mask & SETATTR_GID) {
825 iattr.ia_gid = stbuf->st_gid;
826 iattr.ia_valid |= ATTR_GID;
828 if (mask & SETATTR_LEN) {
829 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
830 iattr.ia_valid |= ATTR_SIZE;
833 iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
834 iattr.ia_ctime = CFS_CURRENT_TIME;
836 rc = llu_setattr_raw(ino, &iattr);
837 liblustre_wait_idle();
841 #define EXT2_LINK_MAX 32000
843 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
845 struct inode *dir = pno->p_base->pb_parent->pb_ino;
846 struct qstr *qstr = &pno->p_base->pb_name;
847 const char *name = qstr->name;
849 struct ptlrpc_request *request = NULL;
850 struct llu_sb_info *sbi = llu_i2sbi(dir);
851 struct md_op_data op_data = {{ 0 }};
855 liblustre_wait_event(0);
856 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
859 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
862 err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
863 S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
864 cfs_curproc_cap_pack(), 0, &request);
865 ptlrpc_req_finished(request);
866 liblustre_wait_event(0);
870 static int llu_readlink_internal(struct inode *inode,
871 struct ptlrpc_request **request,
874 struct llu_inode_info *lli = llu_i2info(inode);
875 struct llu_sb_info *sbi = llu_i2sbi(inode);
876 struct mdt_body *body;
877 struct intnl_stat *st = llu_i2stat(inode);
878 struct md_op_data op_data = {{ 0 }};
879 int rc, symlen = st->st_size + 1;
885 if (lli->lli_symlink_name) {
886 *symname = lli->lli_symlink_name;
887 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
891 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen,
893 op_data.op_valid = OBD_MD_LINKNAME;
895 rc = md_getattr(sbi->ll_md_exp, &op_data, request);
897 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
901 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
902 LASSERT(body != NULL);
904 if ((body->valid & OBD_MD_LINKNAME) == 0) {
905 CERROR ("OBD_MD_LINKNAME not set on reply\n");
906 GOTO (failed, rc = -EPROTO);
909 LASSERT(symlen != 0);
910 if (body->eadatasize != symlen) {
911 CERROR("inode %llu: symlink length %d not expected %d\n",
912 (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
913 GOTO(failed, rc = -EPROTO);
916 *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
917 if (*symname == NULL ||
918 strnlen(*symname, symlen) != symlen - 1) {
919 /* not full/NULL terminated */
920 CERROR("inode %llu: symlink not NULL terminated string"
921 "of length %d\n", (long long)st->st_ino, symlen - 1);
922 GOTO(failed, rc = -EPROTO);
925 OBD_ALLOC(lli->lli_symlink_name, symlen);
926 /* do not return an error if we cannot cache the symlink locally */
927 if (lli->lli_symlink_name)
928 memcpy(lli->lli_symlink_name, *symname, symlen);
933 ptlrpc_req_finished (*request);
937 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
939 struct inode *inode = pno->p_base->pb_ino;
940 struct ptlrpc_request *request;
945 liblustre_wait_event(0);
946 rc = llu_readlink_internal(inode, &request, &symname);
951 strncpy(data, symname, bufsize);
952 rc = strlen(symname);
954 ptlrpc_req_finished(request);
956 liblustre_wait_event(0);
960 static int llu_iop_mknod_raw(struct pnode *pno,
964 struct ptlrpc_request *request = NULL;
965 struct inode *dir = pno->p_parent->p_base->pb_ino;
966 struct llu_sb_info *sbi = llu_i2sbi(dir);
967 struct md_op_data op_data = {{ 0 }};
971 liblustre_wait_event(0);
972 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
973 (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
974 (long long)llu_i2stat(dir)->st_ino);
976 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
979 switch (mode & S_IFMT) {
982 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
987 llu_prep_md_op_data(&op_data, dir, NULL,
988 pno->p_base->pb_name.name,
989 pno->p_base->pb_name.len, 0,
992 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
993 current->fsuid, current->fsgid,
994 cfs_curproc_cap_pack(), dev, &request);
995 ptlrpc_req_finished(request);
1003 liblustre_wait_event(0);
1007 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
1009 struct inode *src = old->p_base->pb_ino;
1010 struct inode *dir = new->p_parent->p_base->pb_ino;
1011 const char *name = new->p_base->pb_name.name;
1012 int namelen = new->p_base->pb_name.len;
1013 struct ptlrpc_request *request = NULL;
1014 struct md_op_data op_data = {{ 0 }};
1021 liblustre_wait_event(0);
1022 llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1024 rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1025 ptlrpc_req_finished(request);
1026 liblustre_wait_event(0);
1032 * libsysio will clear the inode immediately after return
1034 static int llu_iop_unlink_raw(struct pnode *pno)
1036 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1037 struct qstr *qstr = &pno->p_base->pb_name;
1038 const char *name = qstr->name;
1039 int len = qstr->len;
1040 struct inode *target = pno->p_base->pb_ino;
1041 struct ptlrpc_request *request = NULL;
1042 struct md_op_data op_data = { { 0 } };
1048 liblustre_wait_event(0);
1049 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1051 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1053 rc = llu_objects_destroy(request, dir);
1054 ptlrpc_req_finished(request);
1055 liblustre_wait_idle();
1060 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1062 struct inode *src = old->p_parent->p_base->pb_ino;
1063 struct inode *tgt = new->p_parent->p_base->pb_ino;
1064 const char *oldname = old->p_base->pb_name.name;
1065 int oldnamelen = old->p_base->pb_name.len;
1066 const char *newname = new->p_base->pb_name.name;
1067 int newnamelen = new->p_base->pb_name.len;
1068 struct ptlrpc_request *request = NULL;
1069 struct md_op_data op_data = { { 0 } };
1076 liblustre_wait_event(0);
1077 llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1079 rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1080 oldname, oldnamelen, newname, newnamelen,
1083 rc = llu_objects_destroy(request, src);
1086 ptlrpc_req_finished(request);
1087 liblustre_wait_idle();
1092 #ifdef _HAVE_STATVFS
1093 static int llu_statfs_internal(struct llu_sb_info *sbi,
1094 struct obd_statfs *osfs, __u64 max_age)
1096 struct obd_statfs obd_osfs;
1100 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, 0);
1102 CERROR("md_statfs fails: rc = %d\n", rc);
1106 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1107 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1109 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1110 &obd_statfs, max_age, 0);
1112 CERROR("obd_statfs fails: rc = %d\n", rc);
1116 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1117 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1120 osfs->os_blocks = obd_osfs.os_blocks;
1121 osfs->os_bfree = obd_osfs.os_bfree;
1122 osfs->os_bavail = obd_osfs.os_bavail;
1124 /* If we don't have as many objects free on the OST as inodes
1125 * on the MDS, we reduce the total number of inodes to
1126 * compensate, so that the "inodes in use" number is correct.
1128 if (obd_osfs.os_ffree < osfs->os_ffree) {
1129 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1131 osfs->os_ffree = obd_osfs.os_ffree;
1137 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1139 struct obd_statfs osfs;
1142 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1144 /* For now we will always get up-to-date statfs values, but in the
1145 * future we may allow some amount of caching on the client (e.g.
1146 * from QOS or lprocfs updates). */
1147 rc = llu_statfs_internal(sbi, &osfs,
1148 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS));
1152 statfs_unpack(sfs, &osfs);
1154 if (sizeof(sfs->f_blocks) == 4) {
1155 while (osfs.os_blocks > ~0UL) {
1158 osfs.os_blocks >>= 1;
1159 osfs.os_bfree >>= 1;
1160 osfs.os_bavail >>= 1;
1164 sfs->f_blocks = osfs.os_blocks;
1165 sfs->f_bfree = osfs.os_bfree;
1166 sfs->f_bavail = osfs.os_bavail;
1171 static int llu_iop_statvfs(struct pnode *pno,
1173 struct intnl_statvfs *buf)
1179 liblustre_wait_event(0);
1182 LASSERT(pno->p_base->pb_ino);
1183 rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1187 /* from native driver */
1188 buf->f_bsize = fs.f_bsize; /* file system block size */
1189 buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1190 buf->f_blocks = fs.f_blocks;
1191 buf->f_bfree = fs.f_bfree;
1192 buf->f_bavail = fs.f_bavail;
1193 buf->f_files = fs.f_files; /* Total number serial numbers */
1194 buf->f_ffree = fs.f_ffree; /* Number free serial numbers */
1195 buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1196 buf->f_fsid = fs.f_fsid.__val[1];
1197 buf->f_flag = 0; /* No equiv in statfs; maybe use type? */
1198 buf->f_namemax = fs.f_namelen;
1201 liblustre_wait_event(0);
1204 #endif /* _HAVE_STATVFS */
1206 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1208 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1209 struct qstr *qstr = &pno->p_base->pb_name;
1210 const char *name = qstr->name;
1211 int len = qstr->len;
1212 struct ptlrpc_request *request = NULL;
1213 struct intnl_stat *st = llu_i2stat(dir);
1214 struct md_op_data op_data = {{ 0 }};
1218 liblustre_wait_event(0);
1219 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1220 (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1222 if (st->st_nlink >= EXT2_LINK_MAX)
1225 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1228 err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1229 mode | S_IFDIR, current->fsuid, current->fsgid,
1230 cfs_curproc_cap_pack(), 0, &request);
1231 ptlrpc_req_finished(request);
1232 liblustre_wait_event(0);
1236 static int llu_iop_rmdir_raw(struct pnode *pno)
1238 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1239 struct qstr *qstr = &pno->p_base->pb_name;
1240 const char *name = qstr->name;
1241 int len = qstr->len;
1242 struct ptlrpc_request *request = NULL;
1243 struct md_op_data op_data = {{ 0 }};
1247 liblustre_wait_event(0);
1248 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1249 (long long)llu_i2stat(dir)->st_ino,
1250 llu_i2info(dir)->lli_st_generation, dir);
1252 llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1254 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1255 ptlrpc_req_finished(request);
1257 liblustre_wait_event(0);
1262 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1264 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1266 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1268 /* refer to ll_file_flock() for details */
1269 static int llu_file_flock(struct inode *ino,
1271 struct file_lock *file_lock)
1273 struct llu_inode_info *lli = llu_i2info(ino);
1274 struct intnl_stat *st = llu_i2stat(ino);
1275 struct ldlm_res_id res_id =
1276 { .name = {fid_seq(&lli->lli_fid),
1277 fid_oid(&lli->lli_fid),
1278 fid_ver(&lli->lli_fid),
1280 struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
1281 ldlm_flock_completion_ast, NULL, NULL, file_lock };
1283 struct lustre_handle lockh = {0};
1284 ldlm_policy_data_t flock;
1288 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1289 (unsigned long long)st->st_ino, file_lock);
1291 flock.l_flock.pid = file_lock->fl_pid;
1292 flock.l_flock.start = file_lock->fl_start;
1293 flock.l_flock.end = file_lock->fl_end;
1295 switch (file_lock->fl_type) {
1297 einfo.ei_mode = LCK_PR;
1300 einfo.ei_mode = LCK_NL;
1303 einfo.ei_mode = LCK_PW;
1306 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1313 #if F_SETLKW64 != F_SETLKW
1321 #if F_SETLK64 != F_SETLK
1325 flags = LDLM_FL_BLOCK_NOWAIT;
1329 #if F_GETLK64 != F_GETLK
1333 flags = LDLM_FL_TEST_LOCK;
1334 file_lock->fl_type = einfo.ei_mode;
1337 CERROR("unknown fcntl cmd: %d\n", cmd);
1341 CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#x, mode=%u, "
1342 "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1343 flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1347 struct lmv_obd *lmv;
1348 struct obd_device *lmv_obd;
1349 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1350 lmv = &lmv_obd->u.lmv;
1352 if (lmv->desc.ld_tgt_count < 1)
1353 RETURN(rc = -ENODEV);
1355 if (lmv->tgts[0].ltd_exp != NULL)
1356 rc = ldlm_cli_enqueue(lmv->tgts[0].ltd_exp, NULL, &einfo, &res_id,
1357 &flock, &flags, NULL, 0, &lockh, 0);
1364 static int assign_type(struct file_lock *fl, int type)
1377 static int flock_to_posix_lock(struct inode *ino,
1378 struct file_lock *fl,
1381 switch (l->l_whence) {
1382 /* XXX: only SEEK_SET is supported in lustre */
1390 fl->fl_end = l->l_len - 1;
1394 fl->fl_end = OFFSET_MAX;
1396 fl->fl_pid = getpid();
1397 fl->fl_flags = FL_POSIX;
1398 fl->fl_notify = NULL;
1399 fl->fl_insert = NULL;
1400 fl->fl_remove = NULL;
1401 /* XXX: these fields can't be filled with suitable values,
1402 but I think lustre doesn't use them.
1404 fl->fl_owner = NULL;
1407 return assign_type(fl, l->l_type);
1410 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1412 struct file_lock fl;
1416 if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1419 error = flock_to_posix_lock(ino, &fl, flock);
1423 error = llu_file_flock(ino, F_GETLK, &fl);
1427 flock->l_type = F_UNLCK;
1428 if (fl.fl_type != F_UNLCK) {
1429 flock->l_pid = fl.fl_pid;
1430 flock->l_start = fl.fl_start;
1431 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1432 fl.fl_end - fl.fl_start + 1;
1433 flock->l_whence = SEEK_SET;
1434 flock->l_type = fl.fl_type;
1441 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1443 struct file_lock fl;
1444 int flags = llu_i2info(ino)->lli_open_flags + 1;
1447 error = flock_to_posix_lock(ino, &fl, flock);
1450 if (cmd == F_SETLKW)
1451 fl.fl_flags |= FL_SLEEP;
1454 switch (flock->l_type) {
1456 if (!(flags & FMODE_READ))
1460 if (!(flags & FMODE_WRITE))
1470 error = llu_file_flock(ino, cmd, &fl);
1478 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1480 struct llu_inode_info *lli = llu_i2info(ino);
1482 struct flock *flock;
1485 liblustre_wait_event(0);
1488 *rtn = lli->lli_open_flags;
1491 flags = va_arg(ap, long);
1492 flags &= FCNTL_FLMASK;
1493 if (flags & FCNTL_FLMASK_INVALID) {
1494 LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1495 "the O_NONBLOCK or O_ASYNC flags. "
1496 "Please fix your application.\n");
1501 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1502 (lli->lli_open_flags & ~FCNTL_FLMASK);
1507 #if F_GETLK64 != F_GETLK
1511 flock = va_arg(ap, struct flock *);
1512 err = llu_fcntl_getlk(ino, flock);
1517 #if F_SETLKW64 != F_SETLKW
1523 #if F_SETLK64 != F_SETLK
1527 flock = va_arg(ap, struct flock *);
1528 err = llu_fcntl_setlk(ino, cmd, flock);
1532 CERROR("unsupported fcntl cmd %x\n", cmd);
1538 liblustre_wait_event(0);
1542 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1544 struct llu_inode_info *lli = llu_i2info(inode);
1545 struct ll_file_data *fd = lli->lli_file_data;
1547 struct ccc_grouplock grouplock;
1550 if (fd->fd_flags & LL_FILE_IGNORE_LOCK) {
1553 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1556 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1558 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1559 arg, (lli->lli_open_flags & O_NONBLOCK),
1565 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1566 fd->fd_grouplock = grouplock;
1571 int llu_put_grouplock(struct inode *inode, unsigned long arg)
1573 struct llu_inode_info *lli = llu_i2info(inode);
1574 struct ll_file_data *fd = lli->lli_file_data;
1575 struct ccc_grouplock grouplock;
1578 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1581 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1583 if (fd->fd_grouplock.cg_gid != arg)
1586 grouplock = fd->fd_grouplock;
1587 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1588 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1590 cl_put_grouplock(&grouplock);
1595 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1597 struct llu_sb_info *sbi = llu_i2sbi(ino);
1598 struct ptlrpc_request *request = NULL;
1599 struct md_op_data op_data = {{ 0 }};
1600 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1603 llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1606 LASSERT(sizeof(lum) == sizeof(*lump));
1607 LASSERT(sizeof(lum.lmm_objects[0]) ==
1608 sizeof(lump->lmm_objects[0]));
1609 if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1612 switch (lum.lmm_magic) {
1613 case LOV_USER_MAGIC_V1: {
1614 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1615 lustre_swab_lov_user_md_v1(&lum);
1618 case LOV_USER_MAGIC_V3: {
1619 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1620 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1624 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1625 " %#08x != %#08x nor %#08x\n",
1626 lum.lmm_magic, LOV_USER_MAGIC_V1,
1632 /* swabbing is done in lov_setstripe() on server side */
1633 rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1634 sizeof(lum), NULL, 0, &request, NULL);
1636 ptlrpc_req_finished(request);
1637 if (rc != -EPERM && rc != -EACCES)
1638 CERROR("md_setattr fails: rc = %d\n", rc);
1641 ptlrpc_req_finished(request);
1646 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1647 struct lov_user_md *lum, int lum_size)
1649 struct llu_sb_info *sbi = llu_i2sbi(ino);
1650 struct llu_inode_info *lli = llu_i2info(ino);
1651 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1652 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR,
1653 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
1654 struct ptlrpc_request *req = NULL;
1655 struct lustre_md md;
1656 struct md_op_data data = {{ 0 }};
1657 struct lustre_handle lockh;
1662 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1663 PFID(&lli->lli_fid));
1667 llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1669 rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1670 &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1674 req = oit.d.lustre.it_data;
1675 rc = it_open_error(DISP_IT_EXECD, &oit);
1681 rc = it_open_error(DISP_OPEN_OPEN, &oit);
1687 rc = md_get_lustre_md(sbi->ll_md_exp, req,
1688 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1692 llu_update_inode(ino, &md);
1693 llu_local_open(lli, &oit);
1694 /* release intent */
1695 if (lustre_handle_is_used(&lockh))
1696 ldlm_lock_decref(&lockh, LCK_CR);
1697 ptlrpc_req_finished(req);
1699 rc = llu_file_release(ino);
1704 ptlrpc_req_finished(req);
1708 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1710 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1712 int flags = FMODE_WRITE;
1715 LASSERT(sizeof(lum) == sizeof(*lump));
1716 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1717 if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1720 rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1724 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1726 struct intnl_stat *st = llu_i2stat(ino);
1727 if (S_ISREG(st->st_mode))
1728 return llu_lov_file_setstripe(ino, arg);
1729 if (S_ISDIR(st->st_mode))
1730 return llu_lov_dir_setstripe(ino, arg);
1735 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1737 struct lov_stripe_md *lsm = llu_i2info(ino)->lli_smd;
1742 return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1746 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1752 liblustre_wait_event(0);
1755 case LL_IOC_GROUP_LOCK:
1756 arg = va_arg(ap, unsigned long);
1757 rc = llu_get_grouplock(ino, arg);
1759 case LL_IOC_GROUP_UNLOCK:
1760 arg = va_arg(ap, unsigned long);
1761 rc = llu_put_grouplock(ino, arg);
1763 case LL_IOC_LOV_SETSTRIPE:
1764 arg = va_arg(ap, unsigned long);
1765 rc = llu_lov_setstripe(ino, arg);
1767 case LL_IOC_LOV_GETSTRIPE:
1768 arg = va_arg(ap, unsigned long);
1769 rc = llu_lov_getstripe(ino, arg);
1772 CERROR("did not support ioctl cmd %lx\n", request);
1777 liblustre_wait_event(0);
1782 * we already do syncronous read/write
1784 static int llu_iop_sync(struct inode *inode)
1786 liblustre_wait_event(0);
1790 static int llu_iop_datasync(struct inode *inode)
1792 liblustre_wait_event(0);
1796 struct filesys_ops llu_filesys_ops =
1798 fsop_gone: llu_fsop_gone,
1801 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1803 struct inode *inode;
1805 struct file_identifier fileid = {&fid, sizeof(fid)};
1807 if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1808 (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1809 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1811 return ERR_PTR(-EPERM);
1814 /* try to find existing inode */
1815 fid = md->body->fid1;
1817 inode = _sysio_i_find(fs, &fileid);
1819 if (inode->i_zombie/* ||
1820 lli->lli_st_generation != md->body->generation*/) {
1824 llu_update_inode(inode, md);
1829 inode = llu_new_inode(fs, &fid);
1831 llu_update_inode(inode, md);
1837 llu_fsswop_mount(const char *source,
1839 const void *data __IS_UNUSED,
1840 struct pnode *tocover,
1841 struct mount **mntp)
1845 struct pnode_base *rootpb;
1846 struct obd_device *obd;
1847 struct llu_sb_info *sbi;
1848 struct obd_statfs osfs;
1849 static struct qstr noname = { NULL, 0, 0 };
1850 struct ptlrpc_request *request = NULL;
1851 struct lustre_md md;
1853 struct config_llog_instance cfg = {0, };
1854 struct lustre_profile *lprof;
1855 char *zconf_mgsnid, *zconf_profile;
1856 char *osc = NULL, *mdc = NULL;
1857 int async = 1, err = -EINVAL;
1858 struct obd_connect_data ocd = {0,};
1859 struct md_op_data op_data = {{0}};
1860 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1861 const int instlen = sizeof(cfg.cfg_instance) * 2 + 2;
1865 if (ll_parse_mount_target(source,
1868 CERROR("mal-formed target %s\n", source);
1871 if (!zconf_mgsnid || !zconf_profile) {
1872 printf("Liblustre: invalid target %s\n", source);
1875 /* allocate & initialize sbi */
1876 OBD_ALLOC(sbi, sizeof(*sbi));
1880 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
1881 ll_generate_random_uuid(uuid);
1882 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1884 /* generate a string unique to this super, let's try
1885 the address of the super itself.*/
1886 cfg.cfg_instance = sbi;
1888 /* retrive & parse config log */
1889 cfg.cfg_uuid = sbi->ll_sb_uuid;
1890 err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
1892 CERROR("Unable to process log: %s\n", zconf_profile);
1893 GOTO(out_free, err);
1896 lprof = class_get_profile(zconf_profile);
1897 if (lprof == NULL) {
1898 CERROR("No profile found: %s\n", zconf_profile);
1899 GOTO(out_free, err = -EINVAL);
1901 OBD_ALLOC(osc, strlen(lprof->lp_dt) + instlen + 2);
1902 sprintf(osc, "%s-%p", lprof->lp_dt, cfg.cfg_instance);
1904 OBD_ALLOC(mdc, strlen(lprof->lp_md) + instlen + 2);
1905 sprintf(mdc, "%s-%p", lprof->lp_md, cfg.cfg_instance);
1909 GOTO(out_free, err = -EINVAL);
1913 GOTO(out_free, err = -EINVAL);
1916 fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1922 obd = class_name2obd(mdc);
1924 CERROR("MDC %s: not setup or attached\n", mdc);
1925 GOTO(out_free, err = -EINVAL);
1927 obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
1928 sizeof(async), &async, NULL);
1930 ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
1931 OBD_CONNECT_FID | OBD_CONNECT_AT |
1932 OBD_CONNECT_VBR | OBD_CONNECT_FULL20;
1934 #ifdef LIBLUSTRE_POSIX_ACL
1935 ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
1937 ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
1938 ocd.ocd_version = LUSTRE_VERSION_CODE;
1941 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1943 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
1944 GOTO(out_free, err);
1947 err = obd_statfs(obd, &osfs, 100000000, 0);
1952 * FIXME fill fs stat data into sbi here!!! FIXME
1956 obd = class_name2obd(osc);
1958 CERROR("OSC %s: not setup or attached\n", osc);
1959 GOTO(out_md, err = -EINVAL);
1961 obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
1962 sizeof(async), &async, NULL);
1964 obd->obd_upcall.onu_owner = &sbi->ll_lco;
1965 obd->obd_upcall.onu_upcall = cl_ocd_update;
1967 ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
1968 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
1969 OBD_CONNECT_FID | OBD_CONNECT_AT |
1972 ocd.ocd_version = LUSTRE_VERSION_CODE;
1973 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1975 CERROR("cannot connect to %s: rc = %d\n", osc, err);
1978 sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
1979 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
1980 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
1982 fid_zero(&sbi->ll_root_fid);
1983 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
1985 CERROR("cannot mds_connect: rc = %d\n", err);
1986 GOTO(out_lock_cn_cb, err);
1988 if (!fid_is_sane(&sbi->ll_root_fid)) {
1989 CERROR("Invalid root fid during mount\n");
1990 GOTO(out_lock_cn_cb, err = -EINVAL);
1992 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
1994 op_data.op_fid1 = sbi->ll_root_fid;
1995 op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
1996 /* fetch attr of root inode */
1997 err = md_getattr(sbi->ll_md_exp, &op_data, &request);
1999 CERROR("md_getattr failed for root: rc = %d\n", err);
2000 GOTO(out_lock_cn_cb, err);
2003 err = md_get_lustre_md(sbi->ll_md_exp, request,
2004 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2006 CERROR("failed to understand root inode md: rc = %d\n",err);
2007 GOTO(out_request, err);
2010 LASSERT(fid_is_sane(&sbi->ll_root_fid));
2012 root = llu_iget(fs, &md);
2013 if (!root || IS_ERR(root)) {
2014 CERROR("fail to generate root inode\n");
2015 GOTO(out_request, err = -EBADF);
2019 * Generate base path-node for root.
2021 rootpb = _sysio_pb_new(&noname, NULL, root);
2027 err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2029 _sysio_pb_gone(rootpb);
2035 ptlrpc_req_finished(request);
2037 CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2042 _sysio_i_gone(root);
2044 ptlrpc_req_finished(request);
2046 obd_disconnect(sbi->ll_dt_exp);
2048 obd_disconnect(sbi->ll_md_exp);
2051 OBD_FREE(osc, strlen(lprof->lp_dt) + instlen + 2);
2053 OBD_FREE(mdc, strlen(lprof->lp_md) + instlen + 2);
2054 OBD_FREE(sbi, sizeof(*sbi));
2055 liblustre_wait_idle();
2059 struct fssw_ops llu_fssw_ops = {
2063 static struct inode_ops llu_inode_ops = {
2064 inop_lookup: llu_iop_lookup,
2065 inop_getattr: llu_iop_getattr,
2066 inop_setattr: llu_iop_setattr,
2067 inop_filldirentries: llu_iop_filldirentries,
2068 inop_mkdir: llu_iop_mkdir_raw,
2069 inop_rmdir: llu_iop_rmdir_raw,
2070 inop_symlink: llu_iop_symlink_raw,
2071 inop_readlink: llu_iop_readlink,
2072 inop_open: llu_iop_open,
2073 inop_close: llu_iop_close,
2074 inop_link: llu_iop_link_raw,
2075 inop_unlink: llu_iop_unlink_raw,
2076 inop_rename: llu_iop_rename_raw,
2077 inop_pos: llu_iop_pos,
2078 inop_read: llu_iop_read,
2079 inop_write: llu_iop_write,
2080 inop_iodone: llu_iop_iodone,
2081 inop_fcntl: llu_iop_fcntl,
2082 inop_sync: llu_iop_sync,
2083 inop_datasync: llu_iop_datasync,
2084 inop_ioctl: llu_iop_ioctl,
2085 inop_mknod: llu_iop_mknod_raw,
2086 #ifdef _HAVE_STATVFS
2087 inop_statvfs: llu_iop_statvfs,
2089 inop_gone: llu_iop_gone,