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, Whamcloud, Inc.
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_id = lsm->lsm_object_id;
278 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
279 oinfo.oi_oa->o_mode = S_IFREG;
280 oinfo.oi_oa->o_ioepoch = ioepoch;
281 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
282 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
283 OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
284 OBD_MD_FLCTIME | OBD_MD_FLGROUP |
285 OBD_MD_FLATIME | OBD_MD_FLEPOCH;
286 obdo_set_parent_fid(oinfo.oi_oa, &llu_i2info(inode)->lli_fid);
288 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
289 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
292 set = ptlrpc_prep_set();
294 CERROR ("ENOMEM allocing request set\n");
297 rc = obd_getattr_async(llu_i2obdexp(inode), &oinfo, set);
299 rc = ptlrpc_set_wait(set);
300 ptlrpc_set_destroy(set);
302 ccc_inode_lsm_put(inode, lsm);
306 oinfo.oi_oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
307 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
310 obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
311 CDEBUG(D_INODE, "objid "LPX64" size %llu, blocks %llu, "
312 "blksize %llu\n", oinfo.oi_oa->o_id,
313 (long long unsigned)llu_i2stat(inode)->st_size,
314 (long long unsigned)llu_i2stat(inode)->st_blocks,
315 (long long unsigned)llu_i2stat(inode)->st_blksize);
319 static struct inode* llu_new_inode(struct filesys *fs,
323 struct llu_inode_info *lli;
324 struct intnl_stat st = {
327 #ifndef AUTOMOUNT_FILE_NAME
328 .st_mode = fid->f_type & S_IFMT,
330 .st_mode = fid->f_type /* all of the bits! */
333 /* FIXME: fix this later */
340 OBD_ALLOC(lli, sizeof(*lli));
344 /* initialize lli here */
345 lli->lli_sbi = llu_fs2sbi(fs);
346 lli->lli_has_smd = false;
347 lli->lli_symlink_name = NULL;
349 lli->lli_maxbytes = (__u64)(~0UL);
350 lli->lli_file_data = NULL;
352 lli->lli_sysio_fid.fid_data = &lli->lli_fid;
353 lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
356 /* file identifier is needed by functions like _sysio_i_find() */
357 inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
358 &st, 0, &llu_inode_ops, lli);
361 OBD_FREE(lli, sizeof(*lli));
366 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
368 struct lustre_handle lockh;
369 ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
376 fid = &llu_i2info(inode)->lli_fid;
377 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
379 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
380 if (md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
381 LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
387 static int llu_inode_revalidate(struct inode *inode)
389 struct llu_inode_info *lli = llu_i2info(inode);
390 struct intnl_stat *st = llu_i2stat(inode);
394 CERROR("REPORT THIS LINE TO PETER\n");
398 if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
400 struct ptlrpc_request *req = NULL;
401 struct llu_sb_info *sbi = llu_i2sbi(inode);
402 struct md_op_data op_data = { { 0 } };
403 unsigned long valid = OBD_MD_FLGETATTR;
406 /* Why don't we update all valid MDS fields here, if we're
407 * doing an RPC anyways? -phil */
408 if (S_ISREG(st->st_mode)) {
409 ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
410 valid |= OBD_MD_FLEASIZE;
413 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen,
415 op_data.op_valid = valid;
417 rc = md_getattr(sbi->ll_md_exp, &op_data, &req);
419 CERROR("failure %d inode %llu\n", rc,
420 (long long)st->st_ino);
423 rc = md_get_lustre_md(sbi->ll_md_exp, req,
424 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
426 /* XXX Too paranoid? */
427 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
428 !((md.body->valid & OBD_MD_FLNLINK) &&
429 (md.body->nlink == 0))) {
430 CERROR("Asked for %s eadata but got %s (%d)\n",
431 (valid & OBD_MD_FLEASIZE) ? "some" : "no",
432 (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
433 md.body->eadatasize);
436 ptlrpc_req_finished(req);
441 llu_update_inode(inode, &md);
443 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
444 ptlrpc_req_finished(req);
447 if (!lli->lli_has_smd) {
448 /* object not yet allocated, don't validate size */
449 st->st_atime = lli->lli_lvb.lvb_atime;
450 st->st_mtime = lli->lli_lvb.lvb_mtime;
451 st->st_ctime = lli->lli_lvb.lvb_ctime;
455 /* ll_glimpse_size will prefer locally cached writes if they extend
457 RETURN(cl_glimpse_size(inode));
460 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
462 *b = *llu_i2stat(ino);
465 static int llu_iop_getattr(struct pnode *pno,
467 struct intnl_stat *b)
472 liblustre_wait_event(0);
476 LASSERT(pno->p_base->pb_ino);
477 ino = pno->p_base->pb_ino;
479 LASSERT(!pno || pno->p_base->pb_ino == ino);
482 /* libsysio might call us directly without intent lock,
483 * we must re-fetch the attrs here
485 rc = llu_inode_revalidate(ino);
487 copy_stat_buf(ino, b);
488 LASSERT(!llu_i2info(ino)->lli_it);
491 liblustre_wait_event(0);
495 static int null_if_equal(struct ldlm_lock *lock, void *data)
497 if (data == lock->l_ast_data) {
498 lock->l_ast_data = NULL;
500 if (lock->l_req_mode != lock->l_granted_mode)
501 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
504 return LDLM_ITER_CONTINUE;
507 void llu_clear_inode(struct inode *inode)
509 struct llu_inode_info *lli = llu_i2info(inode);
510 struct llu_sb_info *sbi = llu_i2sbi(inode);
511 struct lov_stripe_md *lsm;
514 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
515 (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
518 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
519 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
520 null_if_equal, inode);
522 lsm = ccc_inode_lsm_get(inode);
524 obd_change_cbdata(sbi->ll_dt_exp, lsm, null_if_equal, inode);
525 ccc_inode_lsm_put(inode, lsm);
527 cl_inode_fini(inode);
528 lli->lli_has_smd = false;
530 if (lli->lli_symlink_name) {
531 OBD_FREE(lli->lli_symlink_name,
532 strlen(lli->lli_symlink_name) + 1);
533 lli->lli_symlink_name = NULL;
539 void llu_iop_gone(struct inode *inode)
541 struct llu_inode_info *lli = llu_i2info(inode);
544 liblustre_wait_event(0);
545 llu_clear_inode(inode);
547 OBD_FREE(lli, sizeof(*lli));
551 static int inode_setattr(struct inode * inode, struct iattr * attr)
553 unsigned int ia_valid = attr->ia_valid;
554 struct intnl_stat *st = llu_i2stat(inode);
558 * inode_setattr() is only ever invoked with ATTR_SIZE (by
559 * llu_setattr_raw()) when file has no bodies. Check this.
561 LASSERT(ergo(ia_valid & ATTR_SIZE, !llu_i2info(inode)->lli_has_smd));
563 if (ia_valid & ATTR_SIZE)
564 st->st_size = attr->ia_size;
565 if (ia_valid & ATTR_UID)
566 st->st_uid = attr->ia_uid;
567 if (ia_valid & ATTR_GID)
568 st->st_gid = attr->ia_gid;
569 if (ia_valid & ATTR_ATIME)
570 st->st_atime = attr->ia_atime;
571 if (ia_valid & ATTR_MTIME)
572 st->st_mtime = attr->ia_mtime;
573 if (ia_valid & ATTR_CTIME)
574 st->st_ctime = attr->ia_ctime;
575 if (ia_valid & ATTR_MODE) {
576 st->st_mode = attr->ia_mode;
577 if (!cfs_curproc_is_in_groups(st->st_gid) &&
578 !cfs_capable(CFS_CAP_FSETID))
579 st->st_mode &= ~S_ISGID;
581 /* mark_inode_dirty(inode); */
585 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
586 struct md_open_data **mod)
589 struct llu_sb_info *sbi = llu_i2sbi(inode);
590 struct ptlrpc_request *request = NULL;
594 llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
595 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
599 ptlrpc_req_finished(request);
600 if (rc != -EPERM && rc != -EACCES)
601 CERROR("md_setattr fails: rc = %d\n", rc);
605 rc = md_get_lustre_md(sbi->ll_md_exp, request,
606 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
608 ptlrpc_req_finished(request);
612 /* We call inode_setattr to adjust timestamps.
613 * If there is at least some data in file, we cleared ATTR_SIZE
614 * above to avoid invoking vmtruncate, otherwise it is important
615 * to call vmtruncate in inode_setattr to update inode->i_size
617 inode_setattr(inode, &op_data->op_attr);
618 llu_update_inode(inode, &md);
619 ptlrpc_req_finished(request);
624 /* Close IO epoch and send Size-on-MDS attribute update. */
625 static int llu_setattr_done_writing(struct inode *inode,
626 struct md_op_data *op_data,
627 struct md_open_data *mod)
629 struct llu_inode_info *lli = llu_i2info(inode);
630 struct intnl_stat *st = llu_i2stat(inode);
634 LASSERT(op_data != NULL);
635 if (!S_ISREG(st->st_mode))
638 /* XXX: pass och here for the recovery purpose. */
639 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
640 op_data->op_ioepoch, PFID(&lli->lli_fid));
642 op_data->op_flags = MF_EPOCH_CLOSE;
643 llu_done_writing_attr(inode, op_data);
644 llu_pack_inode2opdata(inode, op_data, NULL);
646 rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
648 /* MDS has instructed us to obtain Size-on-MDS attribute
649 * from OSTs and send setattr to back to MDS. */
650 rc = llu_som_update(inode, op_data);
652 CERROR("inode %llu mdc truncate failed: rc = %d\n",
653 (unsigned long long)st->st_ino, rc);
658 /* If this inode has objects allocated to it (lsm != NULL), then the OST
659 * object(s) determine the file size and mtime. Otherwise, the MDS will
660 * keep these values until such a time that objects are allocated for it.
661 * We do the MDS operations first, as it is checking permissions for us.
662 * We don't to the MDS RPC if there is nothing that we want to store there,
663 * otherwise there is no harm in updating mtime/atime on the MDS if we are
664 * going to do an RPC anyways.
666 * If we are doing a truncate, we will send the mtime and ctime updates
667 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
668 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
671 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
673 int has_lsm = llu_i2info(inode)->lli_has_smd;
674 struct intnl_stat *st = llu_i2stat(inode);
675 int ia_valid = attr->ia_valid;
676 struct md_op_data op_data = { { 0 } };
677 struct md_open_data *mod = NULL;
681 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
683 if (ia_valid & ATTR_SIZE) {
684 if (attr->ia_size > ll_file_maxbytes(inode)) {
685 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
686 (long long)attr->ia_size,
687 ll_file_maxbytes(inode));
691 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
694 /* We mark all of the fields "set" so MDS/OST does not re-set them */
695 if (attr->ia_valid & ATTR_CTIME) {
696 attr->ia_ctime = CFS_CURRENT_TIME;
697 attr->ia_valid |= ATTR_CTIME_SET;
699 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
700 attr->ia_atime = CFS_CURRENT_TIME;
701 attr->ia_valid |= ATTR_ATIME_SET;
703 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
704 attr->ia_mtime = CFS_CURRENT_TIME;
705 attr->ia_valid |= ATTR_MTIME_SET;
708 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
709 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
710 ", now = "CFS_TIME_T"\n",
711 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
712 LTIME_S(CFS_CURRENT_TIME));
714 /* NB: ATTR_SIZE will only be set after this point if the size
715 * resides on the MDS, ie, this file has no objects. */
717 attr->ia_valid &= ~ATTR_SIZE;
719 /* If only OST attributes being set on objects, don't do MDS RPC.
720 * In that case, we need to check permissions and update the local
721 * inode ourselves so we can call obdo_from_inode() always. */
722 if (ia_valid & (has_lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
723 memcpy(&op_data.op_attr, attr, sizeof(*attr));
725 /* Open epoch for truncate. */
726 if (exp_connect_som(llu_i2mdexp(inode)) &&
727 (ia_valid & ATTR_SIZE))
728 op_data.op_flags = MF_EPOCH_OPEN;
729 rc = llu_md_setattr(inode, &op_data, &mod);
733 llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch);
734 if (!has_lsm || !S_ISREG(st->st_mode)) {
735 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
739 /* The OST doesn't check permissions, but the alternative is
740 * a gratuitous RPC to the MDS. We already rely on the client
741 * to do read/write/truncate permission checks, so is mtime OK?
743 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
744 /* from sys_utime() */
745 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
746 if (current->fsuid != st->st_uid &&
747 (rc = ll_permission(inode, MAY_WRITE)) != 0)
750 /* from inode_change_ok() */
751 if (current->fsuid != st->st_uid &&
752 !cfs_capable(CFS_CAP_FOWNER))
758 /* Won't invoke llu_vmtruncate(), as we already cleared
760 inode_setattr(inode, attr);
763 if (ia_valid & ATTR_SIZE)
764 attr->ia_valid |= ATTR_SIZE;
765 if (ia_valid & (ATTR_SIZE |
766 ATTR_ATIME | ATTR_ATIME_SET |
767 ATTR_MTIME | ATTR_MTIME_SET))
768 /* on truncate and utimes send attributes to osts, setting
769 * mtime/atime to past will be performed under PW 0:EOF extent
770 * lock (new_size:EOF for truncate)
771 * it may seem excessive to send mtime/atime updates to osts
772 * when not setting times to past, but it is necessary due to
773 * possible time de-synchronization */
774 rc = cl_setattr_ost(inode, attr, NULL);
777 if (op_data.op_ioepoch)
778 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
779 return rc ? rc : rc1;
782 /* here we simply act as a thin layer to glue it with
783 * llu_setattr_raw(), which is copy from kernel
785 static int llu_iop_setattr(struct pnode *pno,
788 struct intnl_stat *stbuf)
794 liblustre_wait_event(0);
796 LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
797 SETATTR_UID | SETATTR_GID |
798 SETATTR_LEN | SETATTR_MODE)));
799 memset(&iattr, 0, sizeof(iattr));
801 if (mask & SETATTR_MODE) {
802 iattr.ia_mode = stbuf->st_mode;
803 iattr.ia_valid |= ATTR_MODE;
805 if (mask & SETATTR_MTIME) {
806 iattr.ia_mtime = stbuf->st_mtime;
807 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
809 if (mask & SETATTR_ATIME) {
810 iattr.ia_atime = stbuf->st_atime;
811 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
813 if (mask & SETATTR_UID) {
814 iattr.ia_uid = stbuf->st_uid;
815 iattr.ia_valid |= ATTR_UID;
817 if (mask & SETATTR_GID) {
818 iattr.ia_gid = stbuf->st_gid;
819 iattr.ia_valid |= ATTR_GID;
821 if (mask & SETATTR_LEN) {
822 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
823 iattr.ia_valid |= ATTR_SIZE;
826 iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
827 iattr.ia_ctime = CFS_CURRENT_TIME;
829 rc = llu_setattr_raw(ino, &iattr);
830 liblustre_wait_idle();
834 #define EXT2_LINK_MAX 32000
836 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
838 struct inode *dir = pno->p_base->pb_parent->pb_ino;
839 struct qstr *qstr = &pno->p_base->pb_name;
840 const char *name = qstr->name;
842 struct ptlrpc_request *request = NULL;
843 struct llu_sb_info *sbi = llu_i2sbi(dir);
844 struct md_op_data op_data = {{ 0 }};
848 liblustre_wait_event(0);
849 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
852 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
855 err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
856 S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
857 cfs_curproc_cap_pack(), 0, &request);
858 ptlrpc_req_finished(request);
859 liblustre_wait_event(0);
863 static int llu_readlink_internal(struct inode *inode,
864 struct ptlrpc_request **request,
867 struct llu_inode_info *lli = llu_i2info(inode);
868 struct llu_sb_info *sbi = llu_i2sbi(inode);
869 struct mdt_body *body;
870 struct intnl_stat *st = llu_i2stat(inode);
871 struct md_op_data op_data = {{ 0 }};
872 int rc, symlen = st->st_size + 1;
878 if (lli->lli_symlink_name) {
879 *symname = lli->lli_symlink_name;
880 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
884 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen,
886 op_data.op_valid = OBD_MD_LINKNAME;
888 rc = md_getattr(sbi->ll_md_exp, &op_data, request);
890 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
894 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
895 LASSERT(body != NULL);
897 if ((body->valid & OBD_MD_LINKNAME) == 0) {
898 CERROR ("OBD_MD_LINKNAME not set on reply\n");
899 GOTO (failed, rc = -EPROTO);
902 LASSERT(symlen != 0);
903 if (body->eadatasize != symlen) {
904 CERROR("inode %llu: symlink length %d not expected %d\n",
905 (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
906 GOTO(failed, rc = -EPROTO);
909 *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
910 if (*symname == NULL ||
911 strnlen(*symname, symlen) != symlen - 1) {
912 /* not full/NULL terminated */
913 CERROR("inode %llu: symlink not NULL terminated string"
914 "of length %d\n", (long long)st->st_ino, symlen - 1);
915 GOTO(failed, rc = -EPROTO);
918 OBD_ALLOC(lli->lli_symlink_name, symlen);
919 /* do not return an error if we cannot cache the symlink locally */
920 if (lli->lli_symlink_name)
921 memcpy(lli->lli_symlink_name, *symname, symlen);
926 ptlrpc_req_finished (*request);
930 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
932 struct inode *inode = pno->p_base->pb_ino;
933 struct ptlrpc_request *request;
938 liblustre_wait_event(0);
939 rc = llu_readlink_internal(inode, &request, &symname);
944 strncpy(data, symname, bufsize);
945 rc = strlen(symname);
947 ptlrpc_req_finished(request);
949 liblustre_wait_event(0);
953 static int llu_iop_mknod_raw(struct pnode *pno,
957 struct ptlrpc_request *request = NULL;
958 struct inode *dir = pno->p_parent->p_base->pb_ino;
959 struct llu_sb_info *sbi = llu_i2sbi(dir);
960 struct md_op_data op_data = {{ 0 }};
964 liblustre_wait_event(0);
965 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
966 (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
967 (long long)llu_i2stat(dir)->st_ino);
969 if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
972 switch (mode & S_IFMT) {
975 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
980 llu_prep_md_op_data(&op_data, dir, NULL,
981 pno->p_base->pb_name.name,
982 pno->p_base->pb_name.len, 0,
985 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
986 current->fsuid, current->fsgid,
987 cfs_curproc_cap_pack(), dev, &request);
988 ptlrpc_req_finished(request);
996 liblustre_wait_event(0);
1000 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
1002 struct inode *src = old->p_base->pb_ino;
1003 struct inode *dir = new->p_parent->p_base->pb_ino;
1004 const char *name = new->p_base->pb_name.name;
1005 int namelen = new->p_base->pb_name.len;
1006 struct ptlrpc_request *request = NULL;
1007 struct md_op_data op_data = {{ 0 }};
1014 liblustre_wait_event(0);
1015 llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1017 rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1018 ptlrpc_req_finished(request);
1019 liblustre_wait_event(0);
1025 * libsysio will clear the inode immediately after return
1027 static int llu_iop_unlink_raw(struct pnode *pno)
1029 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1030 struct qstr *qstr = &pno->p_base->pb_name;
1031 const char *name = qstr->name;
1032 int len = qstr->len;
1033 struct inode *target = pno->p_base->pb_ino;
1034 struct ptlrpc_request *request = NULL;
1035 struct md_op_data op_data = { { 0 } };
1041 liblustre_wait_event(0);
1042 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1044 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1046 rc = llu_objects_destroy(request, dir);
1047 ptlrpc_req_finished(request);
1048 liblustre_wait_idle();
1053 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1055 struct inode *src = old->p_parent->p_base->pb_ino;
1056 struct inode *tgt = new->p_parent->p_base->pb_ino;
1057 const char *oldname = old->p_base->pb_name.name;
1058 int oldnamelen = old->p_base->pb_name.len;
1059 const char *newname = new->p_base->pb_name.name;
1060 int newnamelen = new->p_base->pb_name.len;
1061 struct ptlrpc_request *request = NULL;
1062 struct md_op_data op_data = { { 0 } };
1069 liblustre_wait_event(0);
1070 llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1072 rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1073 oldname, oldnamelen, newname, newnamelen,
1076 rc = llu_objects_destroy(request, src);
1079 ptlrpc_req_finished(request);
1080 liblustre_wait_idle();
1085 #ifdef _HAVE_STATVFS
1086 static int llu_statfs_internal(struct llu_sb_info *sbi,
1087 struct obd_statfs *osfs, __u64 max_age)
1089 struct obd_statfs obd_osfs;
1093 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, 0);
1095 CERROR("md_statfs fails: rc = %d\n", rc);
1099 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1100 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1102 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1103 &obd_statfs, max_age, 0);
1105 CERROR("obd_statfs fails: rc = %d\n", rc);
1109 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1110 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1113 osfs->os_blocks = obd_osfs.os_blocks;
1114 osfs->os_bfree = obd_osfs.os_bfree;
1115 osfs->os_bavail = obd_osfs.os_bavail;
1117 /* If we don't have as many objects free on the OST as inodes
1118 * on the MDS, we reduce the total number of inodes to
1119 * compensate, so that the "inodes in use" number is correct.
1121 if (obd_osfs.os_ffree < osfs->os_ffree) {
1122 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1124 osfs->os_ffree = obd_osfs.os_ffree;
1130 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1132 struct obd_statfs osfs;
1135 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1137 /* For now we will always get up-to-date statfs values, but in the
1138 * future we may allow some amount of caching on the client (e.g.
1139 * from QOS or lprocfs updates). */
1140 rc = llu_statfs_internal(sbi, &osfs,
1141 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS));
1145 statfs_unpack(sfs, &osfs);
1147 if (sizeof(sfs->f_blocks) == 4) {
1148 while (osfs.os_blocks > ~0UL) {
1151 osfs.os_blocks >>= 1;
1152 osfs.os_bfree >>= 1;
1153 osfs.os_bavail >>= 1;
1157 sfs->f_blocks = osfs.os_blocks;
1158 sfs->f_bfree = osfs.os_bfree;
1159 sfs->f_bavail = osfs.os_bavail;
1164 static int llu_iop_statvfs(struct pnode *pno,
1166 struct intnl_statvfs *buf)
1172 liblustre_wait_event(0);
1175 LASSERT(pno->p_base->pb_ino);
1176 rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1180 /* from native driver */
1181 buf->f_bsize = fs.f_bsize; /* file system block size */
1182 buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1183 buf->f_blocks = fs.f_blocks;
1184 buf->f_bfree = fs.f_bfree;
1185 buf->f_bavail = fs.f_bavail;
1186 buf->f_files = fs.f_files; /* Total number serial numbers */
1187 buf->f_ffree = fs.f_ffree; /* Number free serial numbers */
1188 buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1189 buf->f_fsid = fs.f_fsid.__val[1];
1190 buf->f_flag = 0; /* No equiv in statfs; maybe use type? */
1191 buf->f_namemax = fs.f_namelen;
1194 liblustre_wait_event(0);
1197 #endif /* _HAVE_STATVFS */
1199 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1201 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1202 struct qstr *qstr = &pno->p_base->pb_name;
1203 const char *name = qstr->name;
1204 int len = qstr->len;
1205 struct ptlrpc_request *request = NULL;
1206 struct intnl_stat *st = llu_i2stat(dir);
1207 struct md_op_data op_data = {{ 0 }};
1211 liblustre_wait_event(0);
1212 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1213 (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1215 if (st->st_nlink >= EXT2_LINK_MAX)
1218 llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1221 err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1222 mode | S_IFDIR, current->fsuid, current->fsgid,
1223 cfs_curproc_cap_pack(), 0, &request);
1224 ptlrpc_req_finished(request);
1225 liblustre_wait_event(0);
1229 static int llu_iop_rmdir_raw(struct pnode *pno)
1231 struct inode *dir = pno->p_base->pb_parent->pb_ino;
1232 struct qstr *qstr = &pno->p_base->pb_name;
1233 const char *name = qstr->name;
1234 int len = qstr->len;
1235 struct ptlrpc_request *request = NULL;
1236 struct md_op_data op_data = {{ 0 }};
1240 liblustre_wait_event(0);
1241 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1242 (long long)llu_i2stat(dir)->st_ino,
1243 llu_i2info(dir)->lli_st_generation, dir);
1245 llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1247 rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1248 ptlrpc_req_finished(request);
1250 liblustre_wait_event(0);
1255 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1257 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1259 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1261 /* refer to ll_file_flock() for details */
1262 static int llu_file_flock(struct inode *ino,
1264 struct file_lock *file_lock)
1266 struct llu_inode_info *lli = llu_i2info(ino);
1267 struct intnl_stat *st = llu_i2stat(ino);
1268 struct ldlm_res_id res_id =
1269 { .name = {fid_seq(&lli->lli_fid),
1270 fid_oid(&lli->lli_fid),
1271 fid_ver(&lli->lli_fid),
1273 struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
1274 ldlm_flock_completion_ast, NULL, NULL, file_lock };
1276 struct lustre_handle lockh = {0};
1277 ldlm_policy_data_t flock;
1281 CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1282 (unsigned long long)st->st_ino, file_lock);
1284 flock.l_flock.pid = file_lock->fl_pid;
1285 flock.l_flock.start = file_lock->fl_start;
1286 flock.l_flock.end = file_lock->fl_end;
1288 switch (file_lock->fl_type) {
1290 einfo.ei_mode = LCK_PR;
1293 einfo.ei_mode = LCK_NL;
1296 einfo.ei_mode = LCK_PW;
1299 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1306 #if F_SETLKW64 != F_SETLKW
1314 #if F_SETLK64 != F_SETLK
1318 flags = LDLM_FL_BLOCK_NOWAIT;
1322 #if F_GETLK64 != F_GETLK
1326 flags = LDLM_FL_TEST_LOCK;
1327 file_lock->fl_type = einfo.ei_mode;
1330 CERROR("unknown fcntl cmd: %d\n", cmd);
1334 CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#llx, mode=%u, "
1335 "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1336 flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1340 struct lmv_obd *lmv;
1341 struct obd_device *lmv_obd;
1342 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1343 lmv = &lmv_obd->u.lmv;
1345 if (lmv->desc.ld_tgt_count < 1)
1346 RETURN(rc = -ENODEV);
1348 if (lmv->tgts[0].ltd_exp != NULL)
1349 rc = ldlm_cli_enqueue(lmv->tgts[0].ltd_exp, NULL,
1350 &einfo, &res_id, &flock, &flags,
1351 NULL, 0, LVB_T_NONE, &lockh, 0);
1358 static int assign_type(struct file_lock *fl, int type)
1371 static int flock_to_posix_lock(struct inode *ino,
1372 struct file_lock *fl,
1375 switch (l->l_whence) {
1376 /* XXX: only SEEK_SET is supported in lustre */
1384 fl->fl_end = l->l_len - 1;
1388 fl->fl_end = OFFSET_MAX;
1390 fl->fl_pid = getpid();
1391 fl->fl_flags = FL_POSIX;
1392 fl->fl_notify = NULL;
1393 fl->fl_insert = NULL;
1394 fl->fl_remove = NULL;
1395 /* XXX: these fields can't be filled with suitable values,
1396 but I think lustre doesn't use them.
1398 fl->fl_owner = NULL;
1401 return assign_type(fl, l->l_type);
1404 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1406 struct file_lock fl;
1410 if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1413 error = flock_to_posix_lock(ino, &fl, flock);
1417 error = llu_file_flock(ino, F_GETLK, &fl);
1421 flock->l_type = F_UNLCK;
1422 if (fl.fl_type != F_UNLCK) {
1423 flock->l_pid = fl.fl_pid;
1424 flock->l_start = fl.fl_start;
1425 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1426 fl.fl_end - fl.fl_start + 1;
1427 flock->l_whence = SEEK_SET;
1428 flock->l_type = fl.fl_type;
1435 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1437 struct file_lock fl;
1438 int flags = llu_i2info(ino)->lli_open_flags + 1;
1441 error = flock_to_posix_lock(ino, &fl, flock);
1444 if (cmd == F_SETLKW)
1445 fl.fl_flags |= FL_SLEEP;
1448 switch (flock->l_type) {
1450 if (!(flags & FMODE_READ))
1454 if (!(flags & FMODE_WRITE))
1464 error = llu_file_flock(ino, cmd, &fl);
1472 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1474 struct llu_inode_info *lli = llu_i2info(ino);
1476 struct flock *flock;
1479 liblustre_wait_event(0);
1482 *rtn = lli->lli_open_flags;
1485 flags = va_arg(ap, long);
1486 flags &= FCNTL_FLMASK;
1487 if (flags & FCNTL_FLMASK_INVALID) {
1488 LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1489 "the O_NONBLOCK or O_ASYNC flags. "
1490 "Please fix your application.\n");
1495 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1496 (lli->lli_open_flags & ~FCNTL_FLMASK);
1501 #if F_GETLK64 != F_GETLK
1505 flock = va_arg(ap, struct flock *);
1506 err = llu_fcntl_getlk(ino, flock);
1511 #if F_SETLKW64 != F_SETLKW
1517 #if F_SETLK64 != F_SETLK
1521 flock = va_arg(ap, struct flock *);
1522 err = llu_fcntl_setlk(ino, cmd, flock);
1526 CERROR("unsupported fcntl cmd %x\n", cmd);
1532 liblustre_wait_event(0);
1536 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1538 struct llu_inode_info *lli = llu_i2info(inode);
1539 struct ll_file_data *fd = lli->lli_file_data;
1541 struct ccc_grouplock grouplock;
1544 if (fd->fd_flags & LL_FILE_IGNORE_LOCK) {
1547 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1550 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1552 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1553 arg, (lli->lli_open_flags & O_NONBLOCK),
1559 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1560 fd->fd_grouplock = grouplock;
1565 int llu_put_grouplock(struct inode *inode, unsigned long arg)
1567 struct llu_inode_info *lli = llu_i2info(inode);
1568 struct ll_file_data *fd = lli->lli_file_data;
1569 struct ccc_grouplock grouplock;
1572 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1575 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1577 if (fd->fd_grouplock.cg_gid != arg)
1580 grouplock = fd->fd_grouplock;
1581 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1582 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1584 cl_put_grouplock(&grouplock);
1589 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1591 struct llu_sb_info *sbi = llu_i2sbi(ino);
1592 struct ptlrpc_request *request = NULL;
1593 struct md_op_data op_data = {{ 0 }};
1594 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1597 llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1600 LASSERT(sizeof(lum) == sizeof(*lump));
1601 LASSERT(sizeof(lum.lmm_objects[0]) ==
1602 sizeof(lump->lmm_objects[0]));
1603 if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1606 switch (lum.lmm_magic) {
1607 case LOV_USER_MAGIC_V1: {
1608 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1609 lustre_swab_lov_user_md_v1(&lum);
1612 case LOV_USER_MAGIC_V3: {
1613 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1614 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1618 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1619 " %#08x != %#08x nor %#08x\n",
1620 lum.lmm_magic, LOV_USER_MAGIC_V1,
1626 /* swabbing is done in lov_setstripe() on server side */
1627 rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1628 sizeof(lum), NULL, 0, &request, NULL);
1630 ptlrpc_req_finished(request);
1631 if (rc != -EPERM && rc != -EACCES)
1632 CERROR("md_setattr fails: rc = %d\n", rc);
1635 ptlrpc_req_finished(request);
1640 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1641 struct lov_user_md *lum, int lum_size)
1643 struct llu_sb_info *sbi = llu_i2sbi(ino);
1644 struct llu_inode_info *lli = llu_i2info(ino);
1645 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1646 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR,
1647 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
1648 struct ptlrpc_request *req = NULL;
1649 struct lustre_md md;
1650 struct md_op_data data = {{ 0 }};
1651 struct lustre_handle lockh;
1655 if (lli->lli_has_smd) {
1656 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1657 PFID(&lli->lli_fid));
1661 llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1663 rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1664 &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1668 req = oit.d.lustre.it_data;
1669 rc = it_open_error(DISP_IT_EXECD, &oit);
1675 rc = it_open_error(DISP_OPEN_OPEN, &oit);
1681 rc = md_get_lustre_md(sbi->ll_md_exp, req,
1682 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1686 llu_update_inode(ino, &md);
1687 llu_local_open(lli, &oit);
1688 /* release intent */
1689 if (lustre_handle_is_used(&lockh))
1690 ldlm_lock_decref(&lockh, LCK_CR);
1691 ptlrpc_req_finished(req);
1693 rc = llu_file_release(ino);
1698 ptlrpc_req_finished(req);
1702 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1704 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1706 int flags = FMODE_WRITE;
1709 LASSERT(sizeof(lum) == sizeof(*lump));
1710 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1711 if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1714 rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1718 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1720 struct intnl_stat *st = llu_i2stat(ino);
1721 if (S_ISREG(st->st_mode))
1722 return llu_lov_file_setstripe(ino, arg);
1723 if (S_ISDIR(st->st_mode))
1724 return llu_lov_dir_setstripe(ino, arg);
1729 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1731 struct lov_stripe_md *lsm = NULL;
1734 lsm = ccc_inode_lsm_get(ino);
1736 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1738 ccc_inode_lsm_put(ino, lsm);
1742 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1748 liblustre_wait_event(0);
1751 case LL_IOC_GROUP_LOCK:
1752 arg = va_arg(ap, unsigned long);
1753 rc = llu_get_grouplock(ino, arg);
1755 case LL_IOC_GROUP_UNLOCK:
1756 arg = va_arg(ap, unsigned long);
1757 rc = llu_put_grouplock(ino, arg);
1759 case LL_IOC_LOV_SETSTRIPE:
1760 arg = va_arg(ap, unsigned long);
1761 rc = llu_lov_setstripe(ino, arg);
1763 case LL_IOC_LOV_GETSTRIPE:
1764 arg = va_arg(ap, unsigned long);
1765 rc = llu_lov_getstripe(ino, arg);
1768 CERROR("did not support ioctl cmd %lx\n", request);
1773 liblustre_wait_event(0);
1778 * we already do syncronous read/write
1780 static int llu_iop_sync(struct inode *inode)
1782 liblustre_wait_event(0);
1786 static int llu_iop_datasync(struct inode *inode)
1788 liblustre_wait_event(0);
1792 struct filesys_ops llu_filesys_ops =
1794 fsop_gone: llu_fsop_gone,
1797 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1799 struct inode *inode;
1801 struct file_identifier fileid = {&fid, sizeof(fid)};
1803 if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1804 (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1805 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1807 return ERR_PTR(-EPERM);
1810 /* try to find existing inode */
1811 fid = md->body->fid1;
1813 inode = _sysio_i_find(fs, &fileid);
1815 if (inode->i_zombie/* ||
1816 lli->lli_st_generation != md->body->generation*/) {
1820 llu_update_inode(inode, md);
1825 inode = llu_new_inode(fs, &fid);
1827 llu_update_inode(inode, md);
1833 llu_fsswop_mount(const char *source,
1835 const void *data __IS_UNUSED,
1836 struct pnode *tocover,
1837 struct mount **mntp)
1841 struct pnode_base *rootpb;
1842 struct obd_device *obd;
1843 struct llu_sb_info *sbi;
1844 struct obd_statfs osfs;
1845 static struct qstr noname = { NULL, 0, 0 };
1846 struct ptlrpc_request *request = NULL;
1847 struct lustre_md md;
1849 struct config_llog_instance cfg = {0, };
1850 struct lustre_profile *lprof;
1851 char *zconf_mgsnid, *zconf_profile;
1852 char *osc = NULL, *mdc = NULL;
1853 int async = 1, err = -EINVAL;
1854 struct obd_connect_data ocd = {0,};
1855 struct md_op_data op_data = {{0}};
1856 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1857 const int instlen = sizeof(cfg.cfg_instance) * 2 + 2;
1861 if (ll_parse_mount_target(source,
1864 CERROR("mal-formed target %s\n", source);
1867 if (!zconf_mgsnid || !zconf_profile) {
1868 printf("Liblustre: invalid target %s\n", source);
1871 /* allocate & initialize sbi */
1872 OBD_ALLOC(sbi, sizeof(*sbi));
1876 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
1877 ll_generate_random_uuid(uuid);
1878 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1880 /* generate a string unique to this super, let's try
1881 the address of the super itself.*/
1882 cfg.cfg_instance = sbi;
1884 /* retrive & parse config log */
1885 cfg.cfg_uuid = sbi->ll_sb_uuid;
1886 err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
1888 CERROR("Unable to process log: %s\n", zconf_profile);
1889 GOTO(out_free, err);
1892 lprof = class_get_profile(zconf_profile);
1893 if (lprof == NULL) {
1894 CERROR("No profile found: %s\n", zconf_profile);
1895 GOTO(out_free, err = -EINVAL);
1897 OBD_ALLOC(osc, strlen(lprof->lp_dt) + instlen + 2);
1898 sprintf(osc, "%s-%p", lprof->lp_dt, cfg.cfg_instance);
1900 OBD_ALLOC(mdc, strlen(lprof->lp_md) + instlen + 2);
1901 sprintf(mdc, "%s-%p", lprof->lp_md, cfg.cfg_instance);
1905 GOTO(out_free, err = -EINVAL);
1909 GOTO(out_free, err = -EINVAL);
1912 fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1918 obd = class_name2obd(mdc);
1920 CERROR("MDC %s: not setup or attached\n", mdc);
1921 GOTO(out_free, err = -EINVAL);
1923 obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC),
1924 KEY_ASYNC, sizeof(async), &async, NULL);
1926 ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
1927 OBD_CONNECT_FID | OBD_CONNECT_AT |
1928 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
1929 OBD_CONNECT_LVB_TYPE;
1931 #ifdef LIBLUSTRE_POSIX_ACL
1932 ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
1934 ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
1935 ocd.ocd_version = LUSTRE_VERSION_CODE;
1938 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1940 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
1941 GOTO(out_free, err);
1944 err = obd_statfs(NULL, sbi->ll_md_exp, &osfs, 100000000, 0);
1949 * FIXME fill fs stat data into sbi here!!! FIXME
1953 obd = class_name2obd(osc);
1955 CERROR("OSC %s: not setup or attached\n", osc);
1956 GOTO(out_md, err = -EINVAL);
1958 obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_ASYNC),
1959 KEY_ASYNC, sizeof(async), &async, NULL);
1961 obd->obd_upcall.onu_owner = &sbi->ll_lco;
1962 obd->obd_upcall.onu_upcall = cl_ocd_update;
1964 ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
1965 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
1966 OBD_CONNECT_FID | OBD_CONNECT_AT |
1967 OBD_CONNECT_FULL20 | OBD_CONNECT_EINPROGRESS |
1968 OBD_CONNECT_LVB_TYPE;
1970 ocd.ocd_version = LUSTRE_VERSION_CODE;
1971 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1973 CERROR("cannot connect to %s: rc = %d\n", osc, err);
1976 sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
1977 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
1978 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
1980 fid_zero(&sbi->ll_root_fid);
1981 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
1983 CERROR("cannot mds_connect: rc = %d\n", err);
1984 GOTO(out_lock_cn_cb, err);
1986 if (!fid_is_sane(&sbi->ll_root_fid)) {
1987 CERROR("Invalid root fid during mount\n");
1988 GOTO(out_lock_cn_cb, err = -EINVAL);
1990 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
1992 op_data.op_fid1 = sbi->ll_root_fid;
1993 op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
1994 /* fetch attr of root inode */
1995 err = md_getattr(sbi->ll_md_exp, &op_data, &request);
1997 CERROR("md_getattr failed for root: rc = %d\n", err);
1998 GOTO(out_lock_cn_cb, err);
2001 err = md_get_lustre_md(sbi->ll_md_exp, request,
2002 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2004 CERROR("failed to understand root inode md: rc = %d\n",err);
2005 GOTO(out_request, err);
2008 LASSERT(fid_is_sane(&sbi->ll_root_fid));
2010 root = llu_iget(fs, &md);
2011 if (!root || IS_ERR(root)) {
2012 CERROR("fail to generate root inode\n");
2013 GOTO(out_request, err = -EBADF);
2017 * Generate base path-node for root.
2019 rootpb = _sysio_pb_new(&noname, NULL, root);
2025 err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2027 _sysio_pb_gone(rootpb);
2033 ptlrpc_req_finished(request);
2035 CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2040 _sysio_i_gone(root);
2042 ptlrpc_req_finished(request);
2044 obd_disconnect(sbi->ll_dt_exp);
2046 obd_disconnect(sbi->ll_md_exp);
2049 OBD_FREE(osc, strlen(lprof->lp_dt) + instlen + 2);
2051 OBD_FREE(mdc, strlen(lprof->lp_md) + instlen + 2);
2052 OBD_FREE(sbi, sizeof(*sbi));
2053 liblustre_wait_idle();
2057 struct fssw_ops llu_fssw_ops = {
2061 static struct inode_ops llu_inode_ops = {
2062 inop_lookup: llu_iop_lookup,
2063 inop_getattr: llu_iop_getattr,
2064 inop_setattr: llu_iop_setattr,
2065 inop_filldirentries: llu_iop_filldirentries,
2066 inop_mkdir: llu_iop_mkdir_raw,
2067 inop_rmdir: llu_iop_rmdir_raw,
2068 inop_symlink: llu_iop_symlink_raw,
2069 inop_readlink: llu_iop_readlink,
2070 inop_open: llu_iop_open,
2071 inop_close: llu_iop_close,
2072 inop_link: llu_iop_link_raw,
2073 inop_unlink: llu_iop_unlink_raw,
2074 inop_rename: llu_iop_rename_raw,
2075 inop_pos: llu_iop_pos,
2076 inop_read: llu_iop_read,
2077 inop_write: llu_iop_write,
2078 inop_iodone: llu_iop_iodone,
2079 inop_fcntl: llu_iop_fcntl,
2080 inop_sync: llu_iop_sync,
2081 inop_datasync: llu_iop_datasync,
2082 inop_ioctl: llu_iop_ioctl,
2083 inop_mknod: llu_iop_mknod_raw,
2084 #ifdef _HAVE_STATVFS
2085 inop_statvfs: llu_iop_statvfs,
2087 inop_gone: llu_iop_gone,