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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LLITE
40 #include "../include/lustre_dlm.h"
41 #include "../include/lustre_lite.h"
42 #include <linux/pagemap.h>
43 #include <linux/file.h>
44 #include <linux/mount.h>
45 #include "llite_internal.h"
46 #include "../include/lustre/ll_fiemap.h"
48 #include "../include/cl_object.h"
51 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
53 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
56 static enum llioc_iter
57 ll_iocontrol_call(struct inode *inode, struct file *file,
58 unsigned int cmd, unsigned long arg, int *rcp);
60 static struct ll_file_data *ll_file_data_get(void)
62 struct ll_file_data *fd;
64 fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS);
67 fd->fd_write_failed = false;
71 static void ll_file_data_put(struct ll_file_data *fd)
74 kmem_cache_free(ll_file_data_slab, fd);
77 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
78 struct lustre_handle *fh)
80 op_data->op_fid1 = ll_i2info(inode)->lli_fid;
81 op_data->op_attr.ia_mode = inode->i_mode;
82 op_data->op_attr.ia_atime = inode->i_atime;
83 op_data->op_attr.ia_mtime = inode->i_mtime;
84 op_data->op_attr.ia_ctime = inode->i_ctime;
85 op_data->op_attr.ia_size = i_size_read(inode);
86 op_data->op_attr_blocks = inode->i_blocks;
87 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
88 op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
90 op_data->op_handle = *fh;
92 if (ll_i2info(inode)->lli_flags & LLIF_DATA_MODIFIED)
93 op_data->op_bias |= MDS_DATA_MODIFIED;
97 * Closes the IO epoch and packs all the attributes into @op_data for
100 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
101 struct obd_client_handle *och)
103 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
104 ATTR_MTIME | ATTR_MTIME_SET |
105 ATTR_CTIME | ATTR_CTIME_SET;
107 if (!(och->och_flags & FMODE_WRITE))
110 if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
111 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
113 ll_ioepoch_close(inode, op_data, &och, 0);
116 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
117 ll_prep_md_op_data(op_data, inode, NULL, NULL,
118 0, 0, LUSTRE_OPC_ANY, NULL);
121 static int ll_close_inode_openhandle(struct obd_export *md_exp,
123 struct obd_client_handle *och,
124 const __u64 *data_version)
126 struct obd_export *exp = ll_i2mdexp(inode);
127 struct md_op_data *op_data;
128 struct ptlrpc_request *req = NULL;
129 struct obd_device *obd = class_exp2obd(exp);
135 * XXX: in case of LMV, is this correct to access
138 CERROR("Invalid MDC connection handle %#llx\n",
139 ll_i2mdexp(inode)->exp_handle.h_cookie);
144 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
146 /* XXX We leak openhandle and request here. */
151 ll_prepare_close(inode, op_data, och);
153 /* Pass in data_version implies release. */
154 op_data->op_bias |= MDS_HSM_RELEASE;
155 op_data->op_data_version = *data_version;
156 op_data->op_lease_handle = och->och_lease_handle;
157 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
159 epoch_close = op_data->op_flags & MF_EPOCH_CLOSE;
160 rc = md_close(md_exp, op_data, och->och_mod, &req);
162 /* This close must have the epoch closed. */
163 LASSERT(epoch_close);
164 /* MDS has instructed us to obtain Size-on-MDS attribute from
165 * OSTs and send setattr to back to MDS.
167 rc = ll_som_update(inode, op_data);
169 CERROR("%s: inode "DFID" mdc Size-on-MDS update failed: rc = %d\n",
170 ll_i2mdexp(inode)->exp_obd->obd_name,
171 PFID(ll_inode2fid(inode)), rc);
175 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
176 ll_i2mdexp(inode)->exp_obd->obd_name,
177 PFID(ll_inode2fid(inode)), rc);
180 /* DATA_MODIFIED flag was successfully sent on close, cancel data
183 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
184 struct ll_inode_info *lli = ll_i2info(inode);
186 spin_lock(&lli->lli_lock);
187 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
188 spin_unlock(&lli->lli_lock);
192 rc = ll_objects_destroy(req, inode);
194 CERROR("inode %lu ll_objects destroy: rc = %d\n",
197 if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) {
198 struct mdt_body *body;
200 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
201 if (!(body->valid & OBD_MD_FLRELEASED))
205 ll_finish_md_op_data(op_data);
208 if (exp_connect_som(exp) && !epoch_close &&
209 S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
210 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
212 md_clear_open_replay_data(md_exp, och);
213 /* Free @och if it is not waiting for DONE_WRITING. */
214 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
217 if (req) /* This is close request */
218 ptlrpc_req_finished(req);
222 int ll_md_real_close(struct inode *inode, fmode_t fmode)
224 struct ll_inode_info *lli = ll_i2info(inode);
225 struct obd_client_handle **och_p;
226 struct obd_client_handle *och;
230 if (fmode & FMODE_WRITE) {
231 och_p = &lli->lli_mds_write_och;
232 och_usecount = &lli->lli_open_fd_write_count;
233 } else if (fmode & FMODE_EXEC) {
234 och_p = &lli->lli_mds_exec_och;
235 och_usecount = &lli->lli_open_fd_exec_count;
237 LASSERT(fmode & FMODE_READ);
238 och_p = &lli->lli_mds_read_och;
239 och_usecount = &lli->lli_open_fd_read_count;
242 mutex_lock(&lli->lli_och_mutex);
243 if (*och_usecount > 0) {
244 /* There are still users of this handle, so skip
247 mutex_unlock(&lli->lli_och_mutex);
253 mutex_unlock(&lli->lli_och_mutex);
256 /* There might be a race and this handle may already
259 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
266 static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
269 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
270 struct ll_inode_info *lli = ll_i2info(inode);
272 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
273 struct lustre_handle lockh;
274 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_OPEN} };
277 /* clear group lock, if present */
278 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
279 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
281 if (fd->fd_lease_och) {
284 /* Usually the lease is not released when the
285 * application crashed, we need to release here.
287 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
288 CDEBUG(rc ? D_ERROR : D_INODE,
289 "Clean up lease " DFID " %d/%d\n",
290 PFID(&lli->lli_fid), rc, lease_broken);
292 fd->fd_lease_och = NULL;
296 rc = ll_close_inode_openhandle(md_exp, inode, fd->fd_och, NULL);
301 /* Let's see if we have good enough OPEN lock on the file and if
302 * we can skip talking to MDS
305 mutex_lock(&lli->lli_och_mutex);
306 if (fd->fd_omode & FMODE_WRITE) {
308 LASSERT(lli->lli_open_fd_write_count);
309 lli->lli_open_fd_write_count--;
310 } else if (fd->fd_omode & FMODE_EXEC) {
312 LASSERT(lli->lli_open_fd_exec_count);
313 lli->lli_open_fd_exec_count--;
316 LASSERT(lli->lli_open_fd_read_count);
317 lli->lli_open_fd_read_count--;
319 mutex_unlock(&lli->lli_och_mutex);
321 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
322 LDLM_IBITS, &policy, lockmode, &lockh))
323 rc = ll_md_real_close(inode, fd->fd_omode);
326 LUSTRE_FPRIVATE(file) = NULL;
327 ll_file_data_put(fd);
332 /* While this returns an error code, fput() the caller does not, so we need
333 * to make every effort to clean up all of our state here. Also, applications
334 * rarely check close errors and even if an error is returned they will not
335 * re-try the close call.
337 int ll_file_release(struct inode *inode, struct file *file)
339 struct ll_file_data *fd;
340 struct ll_sb_info *sbi = ll_i2sbi(inode);
341 struct ll_inode_info *lli = ll_i2info(inode);
344 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
345 PFID(ll_inode2fid(inode)), inode);
347 if (!is_root_inode(inode))
348 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
349 fd = LUSTRE_FPRIVATE(file);
352 /* The last ref on @file, maybe not be the owner pid of statahead.
353 * Different processes can open the same dir, "ll_opendir_key" means:
354 * it is me that should stop the statahead thread.
356 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd &&
357 lli->lli_opendir_pid != 0)
358 ll_stop_statahead(inode, lli->lli_opendir_key);
360 if (is_root_inode(inode)) {
361 LUSTRE_FPRIVATE(file) = NULL;
362 ll_file_data_put(fd);
366 if (!S_ISDIR(inode->i_mode)) {
367 lov_read_and_clear_async_rc(lli->lli_clob);
368 lli->lli_async_rc = 0;
371 rc = ll_md_close(sbi->ll_md_exp, inode, file);
373 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
374 libcfs_debug_dumplog();
379 static int ll_intent_file_open(struct dentry *dentry, void *lmm,
380 int lmmsize, struct lookup_intent *itp)
382 struct inode *inode = d_inode(dentry);
383 struct ll_sb_info *sbi = ll_i2sbi(inode);
384 struct dentry *parent = dentry->d_parent;
385 const char *name = dentry->d_name.name;
386 const int len = dentry->d_name.len;
387 struct md_op_data *op_data;
388 struct ptlrpc_request *req;
389 __u32 opc = LUSTRE_OPC_ANY;
392 /* Usually we come here only for NFSD, and we want open lock. */
393 /* We can also get here if there was cached open handle in revalidate_it
394 * but it disappeared while we were getting from there to ll_file_open.
395 * But this means this file was closed and immediately opened which
396 * makes a good candidate for using OPEN lock
398 /* If lmmsize & lmm are not 0, we are just setting stripe info
399 * parameters. No need for the open lock
401 if (!lmm && lmmsize == 0) {
402 struct ll_dentry_data *ldd = ll_d2d(dentry);
404 * If we came via ll_iget_for_nfs, then we need to request
405 * struct ll_dentry_data *ldd = ll_d2d(file->f_dentry);
407 * NB: when ldd is NULL, it must have come via normal
408 * lookup path only, since ll_iget_for_nfs always calls
411 if (ldd && ldd->lld_nfs_dentry) {
412 ldd->lld_nfs_dentry = 0;
413 itp->it_flags |= MDS_OPEN_LOCK;
415 if (itp->it_flags & FMODE_WRITE)
416 opc = LUSTRE_OPC_CREATE;
419 op_data = ll_prep_md_op_data(NULL, d_inode(parent),
423 return PTR_ERR(op_data);
425 itp->it_flags |= MDS_OPEN_BY_FID;
426 rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
427 0 /*unused */, &req, ll_md_blocking_ast, 0);
428 ll_finish_md_op_data(op_data);
430 /* reason for keep own exit path - don`t flood log
431 * with messages with -ESTALE errors.
433 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
434 it_open_error(DISP_OPEN_OPEN, itp))
436 ll_release_openhandle(inode, itp);
440 if (it_disposition(itp, DISP_LOOKUP_NEG)) {
445 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
446 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
447 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
451 rc = ll_prep_inode(&inode, req, NULL, itp);
452 if (!rc && itp->it_lock_mode)
453 ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL);
456 ptlrpc_req_finished(req);
457 ll_intent_drop_lock(itp);
463 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
464 * not believe attributes if a few ioepoch holders exist. Attributes for
465 * previous ioepoch if new one is opened are also skipped by MDS.
467 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
469 if (ioepoch && lli->lli_ioepoch != ioepoch) {
470 lli->lli_ioepoch = ioepoch;
471 CDEBUG(D_INODE, "Epoch %llu opened on "DFID"\n",
472 ioepoch, PFID(&lli->lli_fid));
476 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
477 struct obd_client_handle *och)
479 struct mdt_body *body;
481 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
482 och->och_fh = body->handle;
483 och->och_fid = body->fid1;
484 och->och_lease_handle.cookie = it->it_lock_handle;
485 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
486 och->och_flags = it->it_flags;
488 return md_set_open_replay_data(md_exp, och, it);
491 static int ll_local_open(struct file *file, struct lookup_intent *it,
492 struct ll_file_data *fd, struct obd_client_handle *och)
494 struct inode *inode = file_inode(file);
495 struct ll_inode_info *lli = ll_i2info(inode);
497 LASSERT(!LUSTRE_FPRIVATE(file));
502 struct mdt_body *body;
505 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
509 body = req_capsule_server_get(&it->it_request->rq_pill,
511 ll_ioepoch_open(lli, body->ioepoch);
514 LUSTRE_FPRIVATE(file) = fd;
515 ll_readahead_init(inode, &fd->fd_ras);
516 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
518 /* ll_cl_context initialize */
519 rwlock_init(&fd->fd_lock);
520 INIT_LIST_HEAD(&fd->fd_lccs);
525 /* Open a file, and (for the very first open) create objects on the OSTs at
526 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
527 * creation or open until ll_lov_setstripe() ioctl is called.
529 * If we already have the stripe MD locally then we don't request it in
530 * md_open(), by passing a lmm_size = 0.
532 * It is up to the application to ensure no other processes open this file
533 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
534 * used. We might be able to avoid races of that sort by getting lli_open_sem
535 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
536 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
538 int ll_file_open(struct inode *inode, struct file *file)
540 struct ll_inode_info *lli = ll_i2info(inode);
541 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
542 .it_flags = file->f_flags };
543 struct obd_client_handle **och_p = NULL;
544 __u64 *och_usecount = NULL;
545 struct ll_file_data *fd;
546 int rc = 0, opendir_set = 0;
548 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
549 PFID(ll_inode2fid(inode)), inode, file->f_flags);
551 it = file->private_data; /* XXX: compat macro */
552 file->private_data = NULL; /* prevent ll_local_open assertion */
554 fd = ll_file_data_get();
561 if (S_ISDIR(inode->i_mode)) {
562 spin_lock(&lli->lli_sa_lock);
563 if (!lli->lli_opendir_key && !lli->lli_sai &&
564 lli->lli_opendir_pid == 0) {
565 lli->lli_opendir_key = fd;
566 lli->lli_opendir_pid = current_pid();
569 spin_unlock(&lli->lli_sa_lock);
572 if (is_root_inode(inode)) {
573 LUSTRE_FPRIVATE(file) = fd;
577 if (!it || !it->it_disposition) {
578 /* Convert f_flags into access mode. We cannot use file->f_mode,
579 * because everything but O_ACCMODE mask was stripped from
582 if ((oit.it_flags + 1) & O_ACCMODE)
584 if (file->f_flags & O_TRUNC)
585 oit.it_flags |= FMODE_WRITE;
587 /* kernel only call f_op->open in dentry_open. filp_open calls
588 * dentry_open after call to open_namei that checks permissions.
589 * Only nfsd_open call dentry_open directly without checking
590 * permissions and because of that this code below is safe.
592 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
593 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
595 /* We do not want O_EXCL here, presumably we opened the file
596 * already? XXX - NFS implications?
598 oit.it_flags &= ~O_EXCL;
600 /* bug20584, if "it_flags" contains O_CREAT, the file will be
601 * created if necessary, then "IT_CREAT" should be set to keep
604 if (oit.it_flags & O_CREAT)
605 oit.it_op |= IT_CREAT;
611 /* Let's see if we have file open on MDS already. */
612 if (it->it_flags & FMODE_WRITE) {
613 och_p = &lli->lli_mds_write_och;
614 och_usecount = &lli->lli_open_fd_write_count;
615 } else if (it->it_flags & FMODE_EXEC) {
616 och_p = &lli->lli_mds_exec_och;
617 och_usecount = &lli->lli_open_fd_exec_count;
619 och_p = &lli->lli_mds_read_och;
620 och_usecount = &lli->lli_open_fd_read_count;
623 mutex_lock(&lli->lli_och_mutex);
624 if (*och_p) { /* Open handle is present */
625 if (it_disposition(it, DISP_OPEN_OPEN)) {
626 /* Well, there's extra open request that we do not need,
627 * let's close it somehow. This will decref request.
629 rc = it_open_error(DISP_OPEN_OPEN, it);
631 mutex_unlock(&lli->lli_och_mutex);
635 ll_release_openhandle(inode, it);
639 rc = ll_local_open(file, it, fd, NULL);
642 mutex_unlock(&lli->lli_och_mutex);
646 LASSERT(*och_usecount == 0);
647 if (!it->it_disposition) {
648 /* We cannot just request lock handle now, new ELC code
649 * means that one of other OPEN locks for this file
650 * could be cancelled, and since blocking ast handler
651 * would attempt to grab och_mutex as well, that would
652 * result in a deadlock
654 mutex_unlock(&lli->lli_och_mutex);
655 it->it_create_mode |= M_CHECK_STALE;
656 rc = ll_intent_file_open(file->f_path.dentry, NULL, 0, it);
657 it->it_create_mode &= ~M_CHECK_STALE;
663 *och_p = kzalloc(sizeof(struct obd_client_handle), GFP_NOFS);
671 /* md_intent_lock() didn't get a request ref if there was an
672 * open error, so don't do cleanup on the request here
675 /* XXX (green): Should not we bail out on any error here, not
678 rc = it_open_error(DISP_OPEN_OPEN, it);
682 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
683 "inode %p: disposition %x, status %d\n", inode,
684 it_disposition(it, ~0), it->it_status);
686 rc = ll_local_open(file, it, fd, *och_p);
690 mutex_unlock(&lli->lli_och_mutex);
693 /* Must do this outside lli_och_mutex lock to prevent deadlock where
694 * different kind of OPEN lock for this same inode gets cancelled
697 if (!S_ISREG(inode->i_mode))
700 if (!lli->lli_has_smd &&
701 (cl_is_lov_delay_create(file->f_flags) ||
702 (file->f_mode & FMODE_WRITE) == 0)) {
703 CDEBUG(D_INODE, "object creation was delayed\n");
706 cl_lov_delay_create_clear(&file->f_flags);
711 if (och_p && *och_p) {
716 mutex_unlock(&lli->lli_och_mutex);
719 if (opendir_set != 0)
720 ll_stop_statahead(inode, lli->lli_opendir_key);
721 ll_file_data_put(fd);
723 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
726 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
727 ptlrpc_req_finished(it->it_request);
728 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
734 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
735 struct ldlm_lock_desc *desc,
736 void *data, int flag)
739 struct lustre_handle lockh;
742 case LDLM_CB_BLOCKING:
743 ldlm_lock2handle(lock, &lockh);
744 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
746 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
750 case LDLM_CB_CANCELING:
758 * Acquire a lease and open the file.
760 static struct obd_client_handle *
761 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
764 struct lookup_intent it = { .it_op = IT_OPEN };
765 struct ll_sb_info *sbi = ll_i2sbi(inode);
766 struct md_op_data *op_data;
767 struct ptlrpc_request *req;
768 struct lustre_handle old_handle = { 0 };
769 struct obd_client_handle *och = NULL;
773 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
774 return ERR_PTR(-EINVAL);
777 struct ll_inode_info *lli = ll_i2info(inode);
778 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
779 struct obd_client_handle **och_p;
782 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
783 return ERR_PTR(-EPERM);
785 /* Get the openhandle of the file */
787 mutex_lock(&lli->lli_och_mutex);
788 if (fd->fd_lease_och) {
789 mutex_unlock(&lli->lli_och_mutex);
794 if (file->f_mode & FMODE_WRITE) {
795 LASSERT(lli->lli_mds_write_och);
796 och_p = &lli->lli_mds_write_och;
797 och_usecount = &lli->lli_open_fd_write_count;
799 LASSERT(lli->lli_mds_read_och);
800 och_p = &lli->lli_mds_read_och;
801 och_usecount = &lli->lli_open_fd_read_count;
803 if (*och_usecount == 1) {
810 mutex_unlock(&lli->lli_och_mutex);
811 if (rc < 0) /* more than 1 opener */
815 old_handle = fd->fd_och->och_fh;
818 och = kzalloc(sizeof(*och), GFP_NOFS);
820 return ERR_PTR(-ENOMEM);
822 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
823 LUSTRE_OPC_ANY, NULL);
824 if (IS_ERR(op_data)) {
825 rc = PTR_ERR(op_data);
829 /* To tell the MDT this openhandle is from the same owner */
830 op_data->op_handle = old_handle;
832 it.it_flags = fmode | open_flags;
833 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
834 rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, &req,
835 ll_md_blocking_lease_ast,
836 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
837 * it can be cancelled which may mislead applications that the lease is
839 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
840 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
841 * doesn't deal with openhandle, so normal openhandle will be leaked.
843 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
844 ll_finish_md_op_data(op_data);
845 ptlrpc_req_finished(req);
849 if (it_disposition(&it, DISP_LOOKUP_NEG)) {
854 rc = it_open_error(DISP_OPEN_OPEN, &it);
858 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
859 ll_och_fill(sbi->ll_md_exp, &it, och);
861 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */ {
866 /* already get lease, handle lease lock */
867 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
868 if (it.it_lock_mode == 0 ||
869 it.it_lock_bits != MDS_INODELOCK_OPEN) {
870 /* open lock must return for lease */
871 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
872 PFID(ll_inode2fid(inode)), it.it_lock_mode,
878 ll_intent_release(&it);
882 /* Cancel open lock */
883 if (it.it_lock_mode != 0) {
884 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
887 och->och_lease_handle.cookie = 0ULL;
889 rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, inode, och, NULL);
891 CERROR("%s: error closing file "DFID": %d\n",
892 ll_get_fsname(inode->i_sb, NULL, 0),
893 PFID(&ll_i2info(inode)->lli_fid), rc2);
894 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
896 ll_intent_release(&it);
903 * Release lease and close the file.
904 * It will check if the lease has ever broken.
906 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
909 struct ldlm_lock *lock;
910 bool cancelled = true;
913 lock = ldlm_handle2lock(&och->och_lease_handle);
915 lock_res_and_lock(lock);
916 cancelled = ldlm_is_cancel(lock);
917 unlock_res_and_lock(lock);
921 CDEBUG(D_INODE, "lease for " DFID " broken? %d\n",
922 PFID(&ll_i2info(inode)->lli_fid), cancelled);
925 ldlm_cli_cancel(&och->och_lease_handle, 0);
927 *lease_broken = cancelled;
929 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
934 /* Fills the obdo with the attributes for the lsm */
935 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
936 struct obdo *obdo, __u64 ioepoch, int dv_flags)
938 struct ptlrpc_request_set *set;
939 struct obd_info oinfo = { };
946 oinfo.oi_oa->o_oi = lsm->lsm_oi;
947 oinfo.oi_oa->o_mode = S_IFREG;
948 oinfo.oi_oa->o_ioepoch = ioepoch;
949 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
950 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
951 OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
952 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
953 OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
954 OBD_MD_FLDATAVERSION;
955 if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) {
956 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
957 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
958 if (dv_flags & LL_DV_WR_FLUSH)
959 oinfo.oi_oa->o_flags |= OBD_FL_FLUSH;
962 set = ptlrpc_prep_set();
964 CERROR("cannot allocate ptlrpc set: rc = %d\n", -ENOMEM);
967 rc = obd_getattr_async(exp, &oinfo, set);
969 rc = ptlrpc_set_wait(set);
970 ptlrpc_set_destroy(set);
973 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
974 OBD_MD_FLATIME | OBD_MD_FLMTIME |
975 OBD_MD_FLCTIME | OBD_MD_FLSIZE |
976 OBD_MD_FLDATAVERSION | OBD_MD_FLFLAGS);
977 if (dv_flags & LL_DV_WR_FLUSH &&
978 !(oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
979 oinfo.oi_oa->o_flags & OBD_FL_FLUSH))
986 * Performs the getattr on the inode and updates its fields.
987 * If @sync != 0, perform the getattr under the server-side lock.
989 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
990 __u64 ioepoch, int sync)
992 struct lov_stripe_md *lsm;
995 lsm = ccc_inode_lsm_get(inode);
996 rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode),
997 obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0);
999 struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi;
1001 obdo_refresh_inode(inode, obdo, obdo->o_valid);
1002 CDEBUG(D_INODE, "objid " DOSTID " size %llu, blocks %llu, blksize %lu\n",
1003 POSTID(oi), i_size_read(inode),
1004 (unsigned long long)inode->i_blocks,
1005 1UL << inode->i_blkbits);
1007 ccc_inode_lsm_put(inode, lsm);
1011 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1013 struct ll_inode_info *lli = ll_i2info(inode);
1014 struct cl_object *obj = lli->lli_clob;
1015 struct cl_attr *attr = vvp_env_thread_attr(env);
1021 ll_inode_size_lock(inode);
1023 /* merge timestamps the most recently obtained from mds with
1024 * timestamps obtained from osts
1026 LTIME_S(inode->i_atime) = lli->lli_atime;
1027 LTIME_S(inode->i_mtime) = lli->lli_mtime;
1028 LTIME_S(inode->i_ctime) = lli->lli_ctime;
1030 mtime = LTIME_S(inode->i_mtime);
1031 atime = LTIME_S(inode->i_atime);
1032 ctime = LTIME_S(inode->i_ctime);
1034 cl_object_attr_lock(obj);
1035 rc = cl_object_attr_get(env, obj, attr);
1036 cl_object_attr_unlock(obj);
1039 goto out_size_unlock;
1041 if (atime < attr->cat_atime)
1042 atime = attr->cat_atime;
1044 if (ctime < attr->cat_ctime)
1045 ctime = attr->cat_ctime;
1047 if (mtime < attr->cat_mtime)
1048 mtime = attr->cat_mtime;
1050 CDEBUG(D_VFSTRACE, DFID " updating i_size %llu\n",
1051 PFID(&lli->lli_fid), attr->cat_size);
1053 i_size_write(inode, attr->cat_size);
1055 inode->i_blocks = attr->cat_blocks;
1057 LTIME_S(inode->i_mtime) = mtime;
1058 LTIME_S(inode->i_atime) = atime;
1059 LTIME_S(inode->i_ctime) = ctime;
1062 ll_inode_size_unlock(inode);
1067 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
1070 struct obdo obdo = { 0 };
1073 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, &obdo, 0, 0);
1075 st->st_size = obdo.o_size;
1076 st->st_blocks = obdo.o_blocks;
1077 st->st_mtime = obdo.o_mtime;
1078 st->st_atime = obdo.o_atime;
1079 st->st_ctime = obdo.o_ctime;
1084 static bool file_is_noatime(const struct file *file)
1086 const struct vfsmount *mnt = file->f_path.mnt;
1087 const struct inode *inode = file_inode(file);
1089 /* Adapted from file_accessed() and touch_atime().*/
1090 if (file->f_flags & O_NOATIME)
1093 if (inode->i_flags & S_NOATIME)
1096 if (IS_NOATIME(inode))
1099 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1102 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1105 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1111 void ll_io_init(struct cl_io *io, const struct file *file, int write)
1113 struct inode *inode = file_inode(file);
1115 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1117 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1118 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1119 file->f_flags & O_DIRECT ||
1122 io->ci_obj = ll_i2info(inode)->lli_clob;
1123 io->ci_lockreq = CILR_MAYBE;
1124 if (ll_file_nolock(file)) {
1125 io->ci_lockreq = CILR_NEVER;
1126 io->ci_no_srvlock = 1;
1127 } else if (file->f_flags & O_APPEND) {
1128 io->ci_lockreq = CILR_MANDATORY;
1131 io->ci_noatime = file_is_noatime(file);
1135 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1136 struct file *file, enum cl_io_type iot,
1137 loff_t *ppos, size_t count)
1139 struct ll_inode_info *lli = ll_i2info(file_inode(file));
1140 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1144 CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: %llu, count: %zd\n",
1145 file->f_path.dentry->d_name.name, iot, *ppos, count);
1148 io = vvp_env_thread_io(env);
1149 ll_io_init(io, file, iot == CIT_WRITE);
1151 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1152 struct vvp_io *vio = vvp_env_io(env);
1153 int write_mutex_locked = 0;
1155 vio->vui_fd = LUSTRE_FPRIVATE(file);
1156 vio->vui_io_subtype = args->via_io_subtype;
1158 switch (vio->vui_io_subtype) {
1160 vio->vui_iter = args->u.normal.via_iter;
1161 vio->vui_iocb = args->u.normal.via_iocb;
1162 if ((iot == CIT_WRITE) &&
1163 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1164 if (mutex_lock_interruptible(&lli->
1166 result = -ERESTARTSYS;
1169 write_mutex_locked = 1;
1171 down_read(&lli->lli_trunc_sem);
1174 vio->u.splice.vui_pipe = args->u.splice.via_pipe;
1175 vio->u.splice.vui_flags = args->u.splice.via_flags;
1178 CERROR("Unknown IO type - %u\n", vio->vui_io_subtype);
1181 ll_cl_add(file, env, io);
1182 result = cl_io_loop(env, io);
1183 ll_cl_remove(file, env);
1184 if (args->via_io_subtype == IO_NORMAL)
1185 up_read(&lli->lli_trunc_sem);
1186 if (write_mutex_locked)
1187 mutex_unlock(&lli->lli_write_mutex);
1189 /* cl_io_rw_init() handled IO */
1190 result = io->ci_result;
1193 if (io->ci_nob > 0) {
1194 result = io->ci_nob;
1195 *ppos = io->u.ci_wr.wr.crw_pos;
1199 cl_io_fini(env, io);
1200 /* If any bit been read/written (result != 0), we just return
1201 * short read/write instead of restart io.
1203 if ((result == 0 || result == -ENODATA) && io->ci_need_restart) {
1204 CDEBUG(D_VFSTRACE, "Restart %s on %pD from %lld, count:%zd\n",
1205 iot == CIT_READ ? "read" : "write",
1206 file, *ppos, count);
1207 LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob);
1211 if (iot == CIT_READ) {
1213 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1214 LPROC_LL_READ_BYTES, result);
1215 } else if (iot == CIT_WRITE) {
1217 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1218 LPROC_LL_WRITE_BYTES, result);
1219 fd->fd_write_failed = false;
1220 } else if (result != -ERESTARTSYS) {
1221 fd->fd_write_failed = true;
1224 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1229 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1232 struct vvp_io_args *args;
1236 env = cl_env_get(&refcheck);
1238 return PTR_ERR(env);
1240 args = ll_env_args(env, IO_NORMAL);
1241 args->u.normal.via_iter = to;
1242 args->u.normal.via_iocb = iocb;
1244 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1245 &iocb->ki_pos, iov_iter_count(to));
1246 cl_env_put(env, &refcheck);
1251 * Write to a file (through the page cache).
1253 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1256 struct vvp_io_args *args;
1260 env = cl_env_get(&refcheck);
1262 return PTR_ERR(env);
1264 args = ll_env_args(env, IO_NORMAL);
1265 args->u.normal.via_iter = from;
1266 args->u.normal.via_iocb = iocb;
1268 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1269 &iocb->ki_pos, iov_iter_count(from));
1270 cl_env_put(env, &refcheck);
1275 * Send file content (through pagecache) somewhere with helper
1277 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1278 struct pipe_inode_info *pipe, size_t count,
1282 struct vvp_io_args *args;
1286 env = cl_env_get(&refcheck);
1288 return PTR_ERR(env);
1290 args = ll_env_args(env, IO_SPLICE);
1291 args->u.splice.via_pipe = pipe;
1292 args->u.splice.via_flags = flags;
1294 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1295 cl_env_put(env, &refcheck);
1299 static int ll_lov_recreate(struct inode *inode, struct ost_id *oi, u32 ost_idx)
1301 struct obd_export *exp = ll_i2dtexp(inode);
1302 struct obd_trans_info oti = { 0 };
1303 struct obdo *oa = NULL;
1306 struct lov_stripe_md *lsm = NULL, *lsm2;
1308 oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS);
1312 lsm = ccc_inode_lsm_get(inode);
1313 if (!lsm_has_objects(lsm)) {
1318 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1319 (lsm->lsm_stripe_count));
1321 lsm2 = libcfs_kvzalloc(lsm_size, GFP_NOFS);
1328 oa->o_nlink = ost_idx;
1329 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1330 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1331 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1332 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1333 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
1334 memcpy(lsm2, lsm, lsm_size);
1335 ll_inode_size_lock(inode);
1336 rc = obd_create(NULL, exp, oa, &lsm2, &oti);
1337 ll_inode_size_unlock(inode);
1342 ccc_inode_lsm_put(inode, lsm);
1343 kmem_cache_free(obdo_cachep, oa);
1347 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1349 struct ll_recreate_obj ucreat;
1352 if (!capable(CFS_CAP_SYS_ADMIN))
1355 if (copy_from_user(&ucreat, (struct ll_recreate_obj __user *)arg,
1359 ostid_set_seq_mdt0(&oi);
1360 ostid_set_id(&oi, ucreat.lrc_id);
1361 return ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx);
1364 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1370 if (!capable(CFS_CAP_SYS_ADMIN))
1373 if (copy_from_user(&fid, (struct lu_fid __user *)arg, sizeof(fid)))
1376 fid_to_ostid(&fid, &oi);
1377 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1378 return ll_lov_recreate(inode, &oi, ost_idx);
1381 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1382 __u64 flags, struct lov_user_md *lum,
1385 struct lov_stripe_md *lsm = NULL;
1386 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1389 lsm = ccc_inode_lsm_get(inode);
1391 ccc_inode_lsm_put(inode, lsm);
1392 CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n",
1393 PFID(ll_inode2fid(inode)));
1398 ll_inode_size_lock(inode);
1399 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
1406 ll_release_openhandle(inode, &oit);
1409 ll_inode_size_unlock(inode);
1410 ll_intent_release(&oit);
1411 ccc_inode_lsm_put(inode, lsm);
1415 ptlrpc_req_finished((struct ptlrpc_request *)oit.it_request);
1419 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1420 struct lov_mds_md **lmmp, int *lmm_size,
1421 struct ptlrpc_request **request)
1423 struct ll_sb_info *sbi = ll_i2sbi(inode);
1424 struct mdt_body *body;
1425 struct lov_mds_md *lmm = NULL;
1426 struct ptlrpc_request *req = NULL;
1427 struct md_op_data *op_data;
1430 rc = ll_get_default_mdsize(sbi, &lmmsize);
1434 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1435 strlen(filename), lmmsize,
1436 LUSTRE_OPC_ANY, NULL);
1437 if (IS_ERR(op_data))
1438 return PTR_ERR(op_data);
1440 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1441 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1442 ll_finish_md_op_data(op_data);
1444 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
1449 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1451 lmmsize = body->eadatasize;
1453 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1459 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1461 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1462 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1468 * This is coming from the MDS, so is probably in
1469 * little endian. We convert it to host endian before
1470 * passing it to userspace.
1472 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
1475 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1476 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1479 /* if function called for directory - we should
1480 * avoid swab not existent lsm objects
1482 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1483 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1484 if (S_ISREG(body->mode))
1485 lustre_swab_lov_user_md_objects(
1486 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1488 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1489 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1490 if (S_ISREG(body->mode))
1491 lustre_swab_lov_user_md_objects(
1492 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1499 *lmm_size = lmmsize;
1504 static int ll_lov_setea(struct inode *inode, struct file *file,
1507 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1508 struct lov_user_md *lump;
1509 int lum_size = sizeof(struct lov_user_md) +
1510 sizeof(struct lov_user_ost_data);
1513 if (!capable(CFS_CAP_SYS_ADMIN))
1516 lump = libcfs_kvzalloc(lum_size, GFP_NOFS);
1520 if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) {
1525 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump,
1527 cl_lov_delay_create_clear(&file->f_flags);
1533 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1536 struct lov_user_md_v3 lumv3;
1537 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1538 struct lov_user_md_v1 __user *lumv1p = (void __user *)arg;
1539 struct lov_user_md_v3 __user *lumv3p = (void __user *)arg;
1541 __u64 flags = FMODE_WRITE;
1543 /* first try with v1 which is smaller than v3 */
1544 lum_size = sizeof(struct lov_user_md_v1);
1545 if (copy_from_user(lumv1, lumv1p, lum_size))
1548 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1549 lum_size = sizeof(struct lov_user_md_v3);
1550 if (copy_from_user(&lumv3, lumv3p, lum_size))
1554 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lumv1,
1556 cl_lov_delay_create_clear(&file->f_flags);
1558 struct lov_stripe_md *lsm;
1561 put_user(0, &lumv1p->lmm_stripe_count);
1563 ll_layout_refresh(inode, &gen);
1564 lsm = ccc_inode_lsm_get(inode);
1565 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1566 0, lsm, (void __user *)arg);
1567 ccc_inode_lsm_put(inode, lsm);
1572 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1574 struct lov_stripe_md *lsm;
1577 lsm = ccc_inode_lsm_get(inode);
1579 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
1580 lsm, (void __user *)arg);
1581 ccc_inode_lsm_put(inode, lsm);
1586 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1588 struct ll_inode_info *lli = ll_i2info(inode);
1589 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1590 struct ll_grouplock grouplock;
1594 CWARN("group id for group lock must not be 0\n");
1598 if (ll_file_nolock(file))
1601 spin_lock(&lli->lli_lock);
1602 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1603 CWARN("group lock already existed with gid %lu\n",
1604 fd->fd_grouplock.lg_gid);
1605 spin_unlock(&lli->lli_lock);
1608 LASSERT(!fd->fd_grouplock.lg_lock);
1609 spin_unlock(&lli->lli_lock);
1611 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
1612 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1616 spin_lock(&lli->lli_lock);
1617 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1618 spin_unlock(&lli->lli_lock);
1619 CERROR("another thread just won the race\n");
1620 cl_put_grouplock(&grouplock);
1624 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1625 fd->fd_grouplock = grouplock;
1626 spin_unlock(&lli->lli_lock);
1628 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1632 static int ll_put_grouplock(struct inode *inode, struct file *file,
1635 struct ll_inode_info *lli = ll_i2info(inode);
1636 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1637 struct ll_grouplock grouplock;
1639 spin_lock(&lli->lli_lock);
1640 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1641 spin_unlock(&lli->lli_lock);
1642 CWARN("no group lock held\n");
1645 LASSERT(fd->fd_grouplock.lg_lock);
1647 if (fd->fd_grouplock.lg_gid != arg) {
1648 CWARN("group lock %lu doesn't match current id %lu\n",
1649 arg, fd->fd_grouplock.lg_gid);
1650 spin_unlock(&lli->lli_lock);
1654 grouplock = fd->fd_grouplock;
1655 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1656 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1657 spin_unlock(&lli->lli_lock);
1659 cl_put_grouplock(&grouplock);
1660 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1665 * Close inode open handle
1667 * \param inode [in] inode in question
1668 * \param it [in,out] intent which contains open info and result
1671 * \retval <0 failure
1673 int ll_release_openhandle(struct inode *inode, struct lookup_intent *it)
1675 struct obd_client_handle *och;
1680 /* Root ? Do nothing. */
1681 if (is_root_inode(inode))
1684 /* No open handle to close? Move away */
1685 if (!it_disposition(it, DISP_OPEN_OPEN))
1688 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1690 och = kzalloc(sizeof(*och), GFP_NOFS);
1696 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1698 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1701 /* this one is in place of ll_file_open */
1702 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1703 ptlrpc_req_finished(it->it_request);
1704 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1710 * Get size for inode for which FIEMAP mapping is requested.
1711 * Make the FIEMAP get_info call and returns the result.
1713 static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1716 struct obd_export *exp = ll_i2dtexp(inode);
1717 struct lov_stripe_md *lsm = NULL;
1718 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1719 __u32 vallen = num_bytes;
1722 /* Checks for fiemap flags */
1723 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1724 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1728 /* Check for FIEMAP_FLAG_SYNC */
1729 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1730 rc = filemap_fdatawrite(inode->i_mapping);
1735 lsm = ccc_inode_lsm_get(inode);
1739 /* If the stripe_count > 1 and the application does not understand
1740 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1742 if (lsm->lsm_stripe_count > 1 &&
1743 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1748 fm_key.oa.o_oi = lsm->lsm_oi;
1749 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1751 if (i_size_read(inode) == 0) {
1752 rc = ll_glimpse_size(inode);
1757 obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
1758 obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
1759 /* If filesize is 0, then there would be no objects for mapping */
1760 if (fm_key.oa.o_size == 0) {
1761 fiemap->fm_mapped_extents = 0;
1766 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1768 rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
1771 CERROR("obd_get_info failed: rc = %d\n", rc);
1774 ccc_inode_lsm_put(inode, lsm);
1778 int ll_fid2path(struct inode *inode, void __user *arg)
1780 struct obd_export *exp = ll_i2mdexp(inode);
1781 const struct getinfo_fid2path __user *gfin = arg;
1782 struct getinfo_fid2path *gfout;
1787 if (!capable(CFS_CAP_DAC_READ_SEARCH) &&
1788 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1791 /* Only need to get the buflen */
1792 if (get_user(pathlen, &gfin->gf_pathlen))
1795 if (pathlen > PATH_MAX)
1798 outsize = sizeof(*gfout) + pathlen;
1800 gfout = kzalloc(outsize, GFP_NOFS);
1804 if (copy_from_user(gfout, arg, sizeof(*gfout))) {
1809 /* Call mdc_iocontrol */
1810 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1814 if (copy_to_user(arg, gfout, outsize))
1822 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1824 struct ll_user_fiemap *fiemap_s;
1825 size_t num_bytes, ret_bytes;
1826 unsigned int extent_count;
1829 /* Get the extent count so we can calculate the size of
1830 * required fiemap buffer
1832 if (get_user(extent_count,
1833 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1837 (SIZE_MAX - sizeof(*fiemap_s)) / sizeof(struct ll_fiemap_extent))
1839 num_bytes = sizeof(*fiemap_s) + (extent_count *
1840 sizeof(struct ll_fiemap_extent));
1842 fiemap_s = libcfs_kvzalloc(num_bytes, GFP_NOFS);
1846 /* get the fiemap value */
1847 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1848 sizeof(*fiemap_s))) {
1853 /* If fm_extent_count is non-zero, read the first extent since
1854 * it is used to calculate end_offset and device from previous
1858 if (copy_from_user(&fiemap_s->fm_extents[0],
1859 (char __user *)arg + sizeof(*fiemap_s),
1860 sizeof(struct ll_fiemap_extent))) {
1866 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1870 ret_bytes = sizeof(struct ll_user_fiemap);
1872 if (extent_count != 0)
1873 ret_bytes += (fiemap_s->fm_mapped_extents *
1874 sizeof(struct ll_fiemap_extent));
1876 if (copy_to_user((void __user *)arg, fiemap_s, ret_bytes))
1885 * Read the data_version for inode.
1887 * This value is computed using stripe object version on OST.
1888 * Version is computed using server side locking.
1890 * @param sync if do sync on the OST side;
1892 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1893 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1895 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
1897 struct lov_stripe_md *lsm = NULL;
1898 struct ll_sb_info *sbi = ll_i2sbi(inode);
1899 struct obdo *obdo = NULL;
1902 /* If no stripe, we consider version is 0. */
1903 lsm = ccc_inode_lsm_get(inode);
1904 if (!lsm_has_objects(lsm)) {
1906 CDEBUG(D_INODE, "No object for inode\n");
1911 obdo = kzalloc(sizeof(*obdo), GFP_NOFS);
1917 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, obdo, 0, flags);
1919 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
1922 *data_version = obdo->o_data_version;
1927 ccc_inode_lsm_put(inode, lsm);
1932 * Trigger a HSM release request for the provided inode.
1934 int ll_hsm_release(struct inode *inode)
1936 struct cl_env_nest nest;
1938 struct obd_client_handle *och = NULL;
1939 __u64 data_version = 0;
1942 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
1943 ll_get_fsname(inode->i_sb, NULL, 0),
1944 PFID(&ll_i2info(inode)->lli_fid));
1946 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
1952 /* Grab latest data_version and [am]time values */
1953 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
1957 env = cl_env_nested_get(&nest);
1963 ll_merge_attr(env, inode);
1964 cl_env_nested_put(&nest, env);
1966 /* Release the file.
1967 * NB: lease lock handle is released in mdc_hsm_release_pack() because
1968 * we still need it to pack l_remote_handle to MDT.
1970 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
1975 if (och && !IS_ERR(och)) /* close the file */
1976 ll_lease_close(och, inode, NULL);
1981 struct ll_swap_stack {
1982 struct iattr ia1, ia2;
1984 struct inode *inode1, *inode2;
1985 bool check_dv1, check_dv2;
1988 static int ll_swap_layouts(struct file *file1, struct file *file2,
1989 struct lustre_swap_layouts *lsl)
1991 struct mdc_swap_layouts msl;
1992 struct md_op_data *op_data;
1995 struct ll_swap_stack *llss = NULL;
1998 llss = kzalloc(sizeof(*llss), GFP_NOFS);
2002 llss->inode1 = file_inode(file1);
2003 llss->inode2 = file_inode(file2);
2005 if (!S_ISREG(llss->inode2->i_mode)) {
2010 if (inode_permission(llss->inode1, MAY_WRITE) ||
2011 inode_permission(llss->inode2, MAY_WRITE)) {
2016 if (llss->inode2->i_sb != llss->inode1->i_sb) {
2021 /* we use 2 bool because it is easier to swap than 2 bits */
2022 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2023 llss->check_dv1 = true;
2025 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2026 llss->check_dv2 = true;
2028 /* we cannot use lsl->sl_dvX directly because we may swap them */
2029 llss->dv1 = lsl->sl_dv1;
2030 llss->dv2 = lsl->sl_dv2;
2032 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2033 if (rc == 0) /* same file, done! */ {
2038 if (rc < 0) { /* sequentialize it */
2039 swap(llss->inode1, llss->inode2);
2041 swap(llss->dv1, llss->dv2);
2042 swap(llss->check_dv1, llss->check_dv2);
2046 if (gid != 0) { /* application asks to flush dirty cache */
2047 rc = ll_get_grouplock(llss->inode1, file1, gid);
2051 rc = ll_get_grouplock(llss->inode2, file2, gid);
2053 ll_put_grouplock(llss->inode1, file1, gid);
2058 /* to be able to restore mtime and atime after swap
2059 * we need to first save them
2062 (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) {
2063 llss->ia1.ia_mtime = llss->inode1->i_mtime;
2064 llss->ia1.ia_atime = llss->inode1->i_atime;
2065 llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME;
2066 llss->ia2.ia_mtime = llss->inode2->i_mtime;
2067 llss->ia2.ia_atime = llss->inode2->i_atime;
2068 llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME;
2071 /* ultimate check, before swapping the layouts we check if
2072 * dataversion has changed (if requested)
2074 if (llss->check_dv1) {
2075 rc = ll_data_version(llss->inode1, &dv, 0);
2078 if (dv != llss->dv1) {
2084 if (llss->check_dv2) {
2085 rc = ll_data_version(llss->inode2, &dv, 0);
2088 if (dv != llss->dv2) {
2094 /* struct md_op_data is used to send the swap args to the mdt
2095 * only flags is missing, so we use struct mdc_swap_layouts
2096 * through the md_op_data->op_data
2098 /* flags from user space have to be converted before they are send to
2099 * server, no flag is sent today, they are only used on the client
2103 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2104 0, LUSTRE_OPC_ANY, &msl);
2105 if (IS_ERR(op_data)) {
2106 rc = PTR_ERR(op_data);
2110 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2111 sizeof(*op_data), op_data, NULL);
2112 ll_finish_md_op_data(op_data);
2116 ll_put_grouplock(llss->inode2, file2, gid);
2117 ll_put_grouplock(llss->inode1, file1, gid);
2120 /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */
2124 /* clear useless flags */
2125 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) {
2126 llss->ia1.ia_valid &= ~ATTR_MTIME;
2127 llss->ia2.ia_valid &= ~ATTR_MTIME;
2130 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) {
2131 llss->ia1.ia_valid &= ~ATTR_ATIME;
2132 llss->ia2.ia_valid &= ~ATTR_ATIME;
2135 /* update time if requested */
2137 if (llss->ia2.ia_valid != 0) {
2138 inode_lock(llss->inode1);
2139 rc = ll_setattr(file1->f_path.dentry, &llss->ia2);
2140 inode_unlock(llss->inode1);
2143 if (llss->ia1.ia_valid != 0) {
2146 inode_lock(llss->inode2);
2147 rc1 = ll_setattr(file2->f_path.dentry, &llss->ia1);
2148 inode_unlock(llss->inode2);
2159 static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2161 struct md_op_data *op_data;
2164 /* Detect out-of range masks */
2165 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2168 /* Non-root users are forbidden to set or clear flags which are
2169 * NOT defined in HSM_USER_MASK.
2171 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2172 !capable(CFS_CAP_SYS_ADMIN))
2175 /* Detect out-of range archive id */
2176 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2177 (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE))
2180 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2181 LUSTRE_OPC_ANY, hss);
2182 if (IS_ERR(op_data))
2183 return PTR_ERR(op_data);
2185 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2186 sizeof(*op_data), op_data, NULL);
2188 ll_finish_md_op_data(op_data);
2193 static int ll_hsm_import(struct inode *inode, struct file *file,
2194 struct hsm_user_import *hui)
2196 struct hsm_state_set *hss = NULL;
2197 struct iattr *attr = NULL;
2200 if (!S_ISREG(inode->i_mode))
2204 hss = kzalloc(sizeof(*hss), GFP_NOFS);
2208 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2209 hss->hss_archive_id = hui->hui_archive_id;
2210 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2211 rc = ll_hsm_state_set(inode, hss);
2215 attr = kzalloc(sizeof(*attr), GFP_NOFS);
2221 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2222 attr->ia_mode |= S_IFREG;
2223 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2224 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2225 attr->ia_size = hui->hui_size;
2226 attr->ia_mtime.tv_sec = hui->hui_mtime;
2227 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2228 attr->ia_atime.tv_sec = hui->hui_atime;
2229 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2231 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2232 ATTR_UID | ATTR_GID |
2233 ATTR_MTIME | ATTR_MTIME_SET |
2234 ATTR_ATIME | ATTR_ATIME_SET;
2238 rc = ll_setattr_raw(file->f_path.dentry, attr, true);
2242 inode_unlock(inode);
2251 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2253 struct inode *inode = file_inode(file);
2254 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2257 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),cmd=%x\n",
2258 PFID(ll_inode2fid(inode)), inode, cmd);
2259 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2261 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2262 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2266 case LL_IOC_GETFLAGS:
2267 /* Get the current value of the file flags */
2268 return put_user(fd->fd_flags, (int __user *)arg);
2269 case LL_IOC_SETFLAGS:
2270 case LL_IOC_CLRFLAGS:
2271 /* Set or clear specific file flags */
2272 /* XXX This probably needs checks to ensure the flags are
2273 * not abused, and to handle any flag side effects.
2275 if (get_user(flags, (int __user *)arg))
2278 if (cmd == LL_IOC_SETFLAGS) {
2279 if ((flags & LL_FILE_IGNORE_LOCK) &&
2280 !(file->f_flags & O_DIRECT)) {
2281 CERROR("%s: unable to disable locking on non-O_DIRECT file\n",
2286 fd->fd_flags |= flags;
2288 fd->fd_flags &= ~flags;
2291 case LL_IOC_LOV_SETSTRIPE:
2292 return ll_lov_setstripe(inode, file, arg);
2293 case LL_IOC_LOV_SETEA:
2294 return ll_lov_setea(inode, file, arg);
2295 case LL_IOC_LOV_SWAP_LAYOUTS: {
2297 struct lustre_swap_layouts lsl;
2299 if (copy_from_user(&lsl, (char __user *)arg,
2300 sizeof(struct lustre_swap_layouts)))
2303 if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
2306 file2 = fget(lsl.sl_fd);
2311 if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
2312 rc = ll_swap_layouts(file, file2, &lsl);
2316 case LL_IOC_LOV_GETSTRIPE:
2317 return ll_lov_getstripe(inode, arg);
2318 case LL_IOC_RECREATE_OBJ:
2319 return ll_lov_recreate_obj(inode, arg);
2320 case LL_IOC_RECREATE_FID:
2321 return ll_lov_recreate_fid(inode, arg);
2322 case FSFILT_IOC_FIEMAP:
2323 return ll_ioctl_fiemap(inode, arg);
2324 case FSFILT_IOC_GETFLAGS:
2325 case FSFILT_IOC_SETFLAGS:
2326 return ll_iocontrol(inode, file, cmd, arg);
2327 case FSFILT_IOC_GETVERSION_OLD:
2328 case FSFILT_IOC_GETVERSION:
2329 return put_user(inode->i_generation, (int __user *)arg);
2330 case LL_IOC_GROUP_LOCK:
2331 return ll_get_grouplock(inode, file, arg);
2332 case LL_IOC_GROUP_UNLOCK:
2333 return ll_put_grouplock(inode, file, arg);
2334 case IOC_OBD_STATFS:
2335 return ll_obd_statfs(inode, (void __user *)arg);
2337 /* We need to special case any other ioctls we want to handle,
2338 * to send them to the MDS/OST as appropriate and to properly
2339 * network encode the arg field.
2340 case FSFILT_IOC_SETVERSION_OLD:
2341 case FSFILT_IOC_SETVERSION:
2343 case LL_IOC_FLUSHCTX:
2344 return ll_flush_ctx(inode);
2345 case LL_IOC_PATH2FID: {
2346 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2347 sizeof(struct lu_fid)))
2352 case OBD_IOC_FID2PATH:
2353 return ll_fid2path(inode, (void __user *)arg);
2354 case LL_IOC_DATA_VERSION: {
2355 struct ioc_data_version idv;
2358 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
2361 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2362 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2363 if (rc == 0 && copy_to_user((char __user *)arg, &idv,
2370 case LL_IOC_GET_MDTIDX: {
2373 mdtidx = ll_get_mdt_idx(inode);
2377 if (put_user(mdtidx, (int __user *)arg))
2382 case OBD_IOC_GETDTNAME:
2383 case OBD_IOC_GETMDNAME:
2384 return ll_get_obd_name(inode, cmd, arg);
2385 case LL_IOC_HSM_STATE_GET: {
2386 struct md_op_data *op_data;
2387 struct hsm_user_state *hus;
2390 hus = kzalloc(sizeof(*hus), GFP_NOFS);
2394 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2395 LUSTRE_OPC_ANY, hus);
2396 if (IS_ERR(op_data)) {
2398 return PTR_ERR(op_data);
2401 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2404 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
2407 ll_finish_md_op_data(op_data);
2411 case LL_IOC_HSM_STATE_SET: {
2412 struct hsm_state_set *hss;
2415 hss = memdup_user((char __user *)arg, sizeof(*hss));
2417 return PTR_ERR(hss);
2419 rc = ll_hsm_state_set(inode, hss);
2424 case LL_IOC_HSM_ACTION: {
2425 struct md_op_data *op_data;
2426 struct hsm_current_action *hca;
2429 hca = kzalloc(sizeof(*hca), GFP_NOFS);
2433 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2434 LUSTRE_OPC_ANY, hca);
2435 if (IS_ERR(op_data)) {
2437 return PTR_ERR(op_data);
2440 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2443 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
2446 ll_finish_md_op_data(op_data);
2450 case LL_IOC_SET_LEASE: {
2451 struct ll_inode_info *lli = ll_i2info(inode);
2452 struct obd_client_handle *och = NULL;
2458 if (!(file->f_mode & FMODE_WRITE))
2463 if (!(file->f_mode & FMODE_READ))
2468 mutex_lock(&lli->lli_och_mutex);
2469 if (fd->fd_lease_och) {
2470 och = fd->fd_lease_och;
2471 fd->fd_lease_och = NULL;
2473 mutex_unlock(&lli->lli_och_mutex);
2476 mode = och->och_flags &
2477 (FMODE_READ|FMODE_WRITE);
2478 rc = ll_lease_close(och, inode, &lease_broken);
2479 if (rc == 0 && lease_broken)
2485 /* return the type of lease or error */
2486 return rc < 0 ? rc : (int)mode;
2491 CDEBUG(D_INODE, "Set lease with mode %d\n", mode);
2493 /* apply for lease */
2494 och = ll_lease_open(inode, file, mode, 0);
2496 return PTR_ERR(och);
2499 mutex_lock(&lli->lli_och_mutex);
2500 if (!fd->fd_lease_och) {
2501 fd->fd_lease_och = och;
2504 mutex_unlock(&lli->lli_och_mutex);
2506 /* impossible now that only excl is supported for now */
2507 ll_lease_close(och, inode, &lease_broken);
2512 case LL_IOC_GET_LEASE: {
2513 struct ll_inode_info *lli = ll_i2info(inode);
2514 struct ldlm_lock *lock = NULL;
2517 mutex_lock(&lli->lli_och_mutex);
2518 if (fd->fd_lease_och) {
2519 struct obd_client_handle *och = fd->fd_lease_och;
2521 lock = ldlm_handle2lock(&och->och_lease_handle);
2523 lock_res_and_lock(lock);
2524 if (!ldlm_is_cancel(lock))
2525 rc = och->och_flags &
2526 (FMODE_READ | FMODE_WRITE);
2527 unlock_res_and_lock(lock);
2528 LDLM_LOCK_PUT(lock);
2531 mutex_unlock(&lli->lli_och_mutex);
2534 case LL_IOC_HSM_IMPORT: {
2535 struct hsm_user_import *hui;
2537 hui = memdup_user((void __user *)arg, sizeof(*hui));
2539 return PTR_ERR(hui);
2541 rc = ll_hsm_import(inode, file, hui);
2549 if (ll_iocontrol_call(inode, file, cmd, arg, &err) ==
2553 return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2554 (void __user *)arg);
2559 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2561 struct inode *inode = file_inode(file);
2562 loff_t retval, eof = 0;
2564 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2565 (origin == SEEK_CUR) ? file->f_pos : 0);
2566 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2567 PFID(ll_inode2fid(inode)), inode, retval, retval, origin);
2568 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2570 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2571 retval = ll_glimpse_size(inode);
2574 eof = i_size_read(inode);
2577 retval = generic_file_llseek_size(file, offset, origin,
2578 ll_file_maxbytes(inode), eof);
2582 static int ll_flush(struct file *file, fl_owner_t id)
2584 struct inode *inode = file_inode(file);
2585 struct ll_inode_info *lli = ll_i2info(inode);
2586 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2589 LASSERT(!S_ISDIR(inode->i_mode));
2591 /* catch async errors that were recorded back when async writeback
2592 * failed for pages in this mapping.
2594 rc = lli->lli_async_rc;
2595 lli->lli_async_rc = 0;
2596 err = lov_read_and_clear_async_rc(lli->lli_clob);
2600 /* The application has been told about write failure already.
2601 * Do not report failure again.
2603 if (fd->fd_write_failed)
2605 return rc ? -EIO : 0;
2609 * Called to make sure a portion of file has been written out.
2610 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2612 * Return how many pages have been written.
2614 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2615 enum cl_fsync_mode mode, int ignore_layout)
2617 struct cl_env_nest nest;
2620 struct cl_fsync_io *fio;
2623 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2624 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2627 env = cl_env_nested_get(&nest);
2629 return PTR_ERR(env);
2631 io = vvp_env_thread_io(env);
2632 io->ci_obj = ll_i2info(inode)->lli_clob;
2633 io->ci_ignore_layout = ignore_layout;
2635 /* initialize parameters for sync */
2636 fio = &io->u.ci_fsync;
2637 fio->fi_start = start;
2639 fio->fi_fid = ll_inode2fid(inode);
2640 fio->fi_mode = mode;
2641 fio->fi_nr_written = 0;
2643 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2644 result = cl_io_loop(env, io);
2646 result = io->ci_result;
2648 result = fio->fi_nr_written;
2649 cl_io_fini(env, io);
2650 cl_env_nested_put(&nest, env);
2655 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2657 struct inode *inode = file_inode(file);
2658 struct ll_inode_info *lli = ll_i2info(inode);
2659 struct ptlrpc_request *req;
2662 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2663 PFID(ll_inode2fid(inode)), inode);
2664 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2666 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2669 /* catch async errors that were recorded back when async writeback
2670 * failed for pages in this mapping.
2672 if (!S_ISDIR(inode->i_mode)) {
2673 err = lli->lli_async_rc;
2674 lli->lli_async_rc = 0;
2677 err = lov_read_and_clear_async_rc(lli->lli_clob);
2682 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
2686 ptlrpc_req_finished(req);
2688 if (S_ISREG(inode->i_mode)) {
2689 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2691 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2692 if (rc == 0 && err < 0)
2695 fd->fd_write_failed = true;
2697 fd->fd_write_failed = false;
2700 inode_unlock(inode);
2705 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2707 struct inode *inode = file_inode(file);
2708 struct ll_sb_info *sbi = ll_i2sbi(inode);
2709 struct ldlm_enqueue_info einfo = {
2710 .ei_type = LDLM_FLOCK,
2711 .ei_cb_cp = ldlm_flock_completion_ast,
2712 .ei_cbdata = file_lock,
2714 struct md_op_data *op_data;
2715 struct lustre_handle lockh = {0};
2716 ldlm_policy_data_t flock = { {0} };
2721 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
2722 PFID(ll_inode2fid(inode)), file_lock);
2724 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2726 if (file_lock->fl_flags & FL_FLOCK)
2727 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2728 else if (!(file_lock->fl_flags & FL_POSIX))
2731 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2732 flock.l_flock.pid = file_lock->fl_pid;
2733 flock.l_flock.start = file_lock->fl_start;
2734 flock.l_flock.end = file_lock->fl_end;
2736 /* Somewhat ugly workaround for svc lockd.
2737 * lockd installs custom fl_lmops->lm_compare_owner that checks
2738 * for the fl_owner to be the same (which it always is on local node
2739 * I guess between lockd processes) and then compares pid.
2740 * As such we assign pid to the owner field to make it all work,
2741 * conflict with normal locks is unlikely since pid space and
2742 * pointer space for current->files are not intersecting
2744 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2745 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2747 switch (file_lock->fl_type) {
2749 einfo.ei_mode = LCK_PR;
2752 /* An unlock request may or may not have any relation to
2753 * existing locks so we may not be able to pass a lock handle
2754 * via a normal ldlm_lock_cancel() request. The request may even
2755 * unlock a byte range in the middle of an existing lock. In
2756 * order to process an unlock request we need all of the same
2757 * information that is given with a normal read or write record
2758 * lock request. To avoid creating another ldlm unlock (cancel)
2759 * message we'll treat a LCK_NL flock request as an unlock.
2761 einfo.ei_mode = LCK_NL;
2764 einfo.ei_mode = LCK_PW;
2767 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n",
2768 file_lock->fl_type);
2783 flags = LDLM_FL_BLOCK_NOWAIT;
2789 flags = LDLM_FL_TEST_LOCK;
2790 /* Save the old mode so that if the mode in the lock changes we
2791 * can decrement the appropriate reader or writer refcount.
2793 file_lock->fl_type = einfo.ei_mode;
2796 CERROR("unknown fcntl lock command: %d\n", cmd);
2800 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2801 LUSTRE_OPC_ANY, NULL);
2802 if (IS_ERR(op_data))
2803 return PTR_ERR(op_data);
2805 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n",
2806 PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags,
2807 einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2809 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2810 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2812 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
2813 !(flags & LDLM_FL_TEST_LOCK))
2814 rc2 = locks_lock_file_wait(file, file_lock);
2816 if (rc2 && file_lock->fl_type != F_UNLCK) {
2817 einfo.ei_mode = LCK_NL;
2818 md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2819 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2823 ll_finish_md_op_data(op_data);
2829 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2835 * test if some locks matching bits and l_req_mode are acquired
2836 * - bits can be in different locks
2837 * - if found clear the common lock bits in *bits
2838 * - the bits not found, are kept in *bits
2840 * \param bits [IN] searched lock bits [IN]
2841 * \param l_req_mode [IN] searched lock mode
2842 * \retval boolean, true iff all bits are found
2844 int ll_have_md_lock(struct inode *inode, __u64 *bits,
2845 enum ldlm_mode l_req_mode)
2847 struct lustre_handle lockh;
2848 ldlm_policy_data_t policy;
2849 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
2850 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
2858 fid = &ll_i2info(inode)->lli_fid;
2859 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
2860 ldlm_lockname[mode]);
2862 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2863 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
2864 policy.l_inodebits.bits = *bits & (1 << i);
2865 if (policy.l_inodebits.bits == 0)
2868 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
2869 &policy, mode, &lockh)) {
2870 struct ldlm_lock *lock;
2872 lock = ldlm_handle2lock(&lockh);
2875 ~(lock->l_policy_data.l_inodebits.bits);
2876 LDLM_LOCK_PUT(lock);
2878 *bits &= ~policy.l_inodebits.bits;
2885 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
2886 struct lustre_handle *lockh, __u64 flags,
2887 enum ldlm_mode mode)
2889 ldlm_policy_data_t policy = { .l_inodebits = {bits} };
2893 fid = &ll_i2info(inode)->lli_fid;
2894 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2896 rc = md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED,
2897 fid, LDLM_IBITS, &policy, mode, lockh);
2902 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
2904 /* Already unlinked. Just update nlink and return success */
2905 if (rc == -ENOENT) {
2907 /* This path cannot be hit for regular files unless in
2908 * case of obscure races, so no need to validate size.
2910 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2912 } else if (rc != 0) {
2913 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
2914 "%s: revalidate FID "DFID" error: rc = %d\n",
2915 ll_get_fsname(inode->i_sb, NULL, 0),
2916 PFID(ll_inode2fid(inode)), rc);
2922 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2924 struct inode *inode = d_inode(dentry);
2925 struct ptlrpc_request *req = NULL;
2926 struct obd_export *exp;
2929 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%pd\n",
2930 PFID(ll_inode2fid(inode)), inode, dentry);
2932 exp = ll_i2mdexp(inode);
2934 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2935 * But under CMD case, it caused some lock issues, should be fixed
2936 * with new CMD ibits lock. See bug 12718
2938 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
2939 struct lookup_intent oit = { .it_op = IT_GETATTR };
2940 struct md_op_data *op_data;
2942 if (ibits == MDS_INODELOCK_LOOKUP)
2943 oit.it_op = IT_LOOKUP;
2945 /* Call getattr by fid, so do not provide name at all. */
2946 op_data = ll_prep_md_op_data(NULL, inode,
2948 LUSTRE_OPC_ANY, NULL);
2949 if (IS_ERR(op_data))
2950 return PTR_ERR(op_data);
2952 oit.it_create_mode |= M_CHECK_STALE;
2953 rc = md_intent_lock(exp, op_data, NULL, 0,
2954 /* we are not interested in name
2958 ll_md_blocking_ast, 0);
2959 ll_finish_md_op_data(op_data);
2960 oit.it_create_mode &= ~M_CHECK_STALE;
2962 rc = ll_inode_revalidate_fini(inode, rc);
2966 rc = ll_revalidate_it_finish(req, &oit, inode);
2968 ll_intent_release(&oit);
2972 /* Unlinked? Unhash dentry, so it is not picked up later by
2973 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2974 * here to preserve get_cwd functionality on 2.6.
2977 if (!d_inode(dentry)->i_nlink) {
2978 spin_lock(&inode->i_lock);
2979 d_lustre_invalidate(dentry, 0);
2980 spin_unlock(&inode->i_lock);
2983 ll_lookup_finish_locks(&oit, inode);
2984 } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) {
2985 struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry));
2986 u64 valid = OBD_MD_FLGETATTR;
2987 struct md_op_data *op_data;
2990 if (S_ISREG(inode->i_mode)) {
2991 rc = ll_get_default_mdsize(sbi, &ealen);
2994 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2997 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2998 0, ealen, LUSTRE_OPC_ANY,
3000 if (IS_ERR(op_data))
3001 return PTR_ERR(op_data);
3003 op_data->op_valid = valid;
3004 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
3005 ll_finish_md_op_data(op_data);
3007 rc = ll_inode_revalidate_fini(inode, rc);
3011 rc = ll_prep_inode(&inode, req, NULL, NULL);
3014 ptlrpc_req_finished(req);
3018 static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3020 struct inode *inode = d_inode(dentry);
3023 rc = __ll_inode_revalidate(dentry, ibits);
3027 /* if object isn't regular file, don't validate size */
3028 if (!S_ISREG(inode->i_mode)) {
3029 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime;
3030 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime;
3031 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime;
3033 /* In case of restore, the MDT has the right size and has
3034 * already send it back without granting the layout lock,
3035 * inode is up-to-date so glimpse is useless.
3036 * Also to glimpse we need the layout, in case of a running
3037 * restore the MDT holds the layout lock so the glimpse will
3038 * block up to the end of restore (getattr will block)
3040 if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING))
3041 rc = ll_glimpse_size(inode);
3046 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3048 struct inode *inode = d_inode(de);
3049 struct ll_sb_info *sbi = ll_i2sbi(inode);
3050 struct ll_inode_info *lli = ll_i2info(inode);
3053 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3054 MDS_INODELOCK_LOOKUP);
3055 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3060 stat->dev = inode->i_sb->s_dev;
3061 if (ll_need_32bit_api(sbi))
3062 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3064 stat->ino = inode->i_ino;
3065 stat->mode = inode->i_mode;
3066 stat->nlink = inode->i_nlink;
3067 stat->uid = inode->i_uid;
3068 stat->gid = inode->i_gid;
3069 stat->rdev = inode->i_rdev;
3070 stat->atime = inode->i_atime;
3071 stat->mtime = inode->i_mtime;
3072 stat->ctime = inode->i_ctime;
3073 stat->blksize = 1 << inode->i_blkbits;
3075 stat->size = i_size_read(inode);
3076 stat->blocks = inode->i_blocks;
3081 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3082 __u64 start, __u64 len)
3086 struct ll_user_fiemap *fiemap;
3087 unsigned int extent_count = fieinfo->fi_extents_max;
3089 num_bytes = sizeof(*fiemap) + (extent_count *
3090 sizeof(struct ll_fiemap_extent));
3091 fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS);
3096 fiemap->fm_flags = fieinfo->fi_flags;
3097 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3098 fiemap->fm_start = start;
3099 fiemap->fm_length = len;
3100 if (extent_count > 0 &&
3101 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3102 sizeof(struct ll_fiemap_extent)) != 0) {
3107 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3109 fieinfo->fi_flags = fiemap->fm_flags;
3110 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3111 if (extent_count > 0 &&
3112 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3113 fiemap->fm_mapped_extents *
3114 sizeof(struct ll_fiemap_extent)) != 0) {
3124 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3126 struct ll_inode_info *lli = ll_i2info(inode);
3127 struct posix_acl *acl = NULL;
3129 spin_lock(&lli->lli_lock);
3130 /* VFS' acl_permission_check->check_acl will release the refcount */
3131 acl = posix_acl_dup(lli->lli_posix_acl);
3132 #ifdef CONFIG_FS_POSIX_ACL
3133 forget_cached_acl(inode, type);
3135 spin_unlock(&lli->lli_lock);
3140 int ll_inode_permission(struct inode *inode, int mask)
3144 if (mask & MAY_NOT_BLOCK)
3147 /* as root inode are NOT getting validated in lookup operation,
3148 * need to do it before permission check.
3151 if (is_root_inode(inode)) {
3152 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3153 MDS_INODELOCK_LOOKUP);
3158 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3159 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3161 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
3162 rc = generic_permission(inode, mask);
3167 /* -o localflock - only provides locally consistent flock locks */
3168 struct file_operations ll_file_operations = {
3169 .read_iter = ll_file_read_iter,
3170 .write_iter = ll_file_write_iter,
3171 .unlocked_ioctl = ll_file_ioctl,
3172 .open = ll_file_open,
3173 .release = ll_file_release,
3174 .mmap = ll_file_mmap,
3175 .llseek = ll_file_seek,
3176 .splice_read = ll_file_splice_read,
3181 struct file_operations ll_file_operations_flock = {
3182 .read_iter = ll_file_read_iter,
3183 .write_iter = ll_file_write_iter,
3184 .unlocked_ioctl = ll_file_ioctl,
3185 .open = ll_file_open,
3186 .release = ll_file_release,
3187 .mmap = ll_file_mmap,
3188 .llseek = ll_file_seek,
3189 .splice_read = ll_file_splice_read,
3192 .flock = ll_file_flock,
3193 .lock = ll_file_flock
3196 /* These are for -o noflock - to return ENOSYS on flock calls */
3197 struct file_operations ll_file_operations_noflock = {
3198 .read_iter = ll_file_read_iter,
3199 .write_iter = ll_file_write_iter,
3200 .unlocked_ioctl = ll_file_ioctl,
3201 .open = ll_file_open,
3202 .release = ll_file_release,
3203 .mmap = ll_file_mmap,
3204 .llseek = ll_file_seek,
3205 .splice_read = ll_file_splice_read,
3208 .flock = ll_file_noflock,
3209 .lock = ll_file_noflock
3212 const struct inode_operations ll_file_inode_operations = {
3213 .setattr = ll_setattr,
3214 .getattr = ll_getattr,
3215 .permission = ll_inode_permission,
3216 .setxattr = ll_setxattr,
3217 .getxattr = ll_getxattr,
3218 .listxattr = ll_listxattr,
3219 .removexattr = ll_removexattr,
3220 .fiemap = ll_fiemap,
3221 .get_acl = ll_get_acl,
3224 /* dynamic ioctl number support routines */
3225 static struct llioc_ctl_data {
3226 struct rw_semaphore ioc_sem;
3227 struct list_head ioc_head;
3229 __RWSEM_INITIALIZER(llioc.ioc_sem),
3230 LIST_HEAD_INIT(llioc.ioc_head)
3234 struct list_head iocd_list;
3235 unsigned int iocd_size;
3236 llioc_callback_t iocd_cb;
3237 unsigned int iocd_count;
3238 unsigned int iocd_cmd[0];
3241 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3244 struct llioc_data *in_data = NULL;
3246 if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0)
3249 size = sizeof(*in_data) + count * sizeof(unsigned int);
3250 in_data = kzalloc(size, GFP_NOFS);
3254 memset(in_data, 0, sizeof(*in_data));
3255 in_data->iocd_size = size;
3256 in_data->iocd_cb = cb;
3257 in_data->iocd_count = count;
3258 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3260 down_write(&llioc.ioc_sem);
3261 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3262 up_write(&llioc.ioc_sem);
3266 EXPORT_SYMBOL(ll_iocontrol_register);
3268 void ll_iocontrol_unregister(void *magic)
3270 struct llioc_data *tmp;
3275 down_write(&llioc.ioc_sem);
3276 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3278 list_del(&tmp->iocd_list);
3279 up_write(&llioc.ioc_sem);
3285 up_write(&llioc.ioc_sem);
3287 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3289 EXPORT_SYMBOL(ll_iocontrol_unregister);
3291 static enum llioc_iter
3292 ll_iocontrol_call(struct inode *inode, struct file *file,
3293 unsigned int cmd, unsigned long arg, int *rcp)
3295 enum llioc_iter ret = LLIOC_CONT;
3296 struct llioc_data *data;
3297 int rc = -EINVAL, i;
3299 down_read(&llioc.ioc_sem);
3300 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3301 for (i = 0; i < data->iocd_count; i++) {
3302 if (cmd != data->iocd_cmd[i])
3305 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3309 if (ret == LLIOC_STOP)
3312 up_read(&llioc.ioc_sem);
3319 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3321 struct ll_inode_info *lli = ll_i2info(inode);
3322 struct cl_env_nest nest;
3329 env = cl_env_nested_get(&nest);
3331 return PTR_ERR(env);
3333 result = cl_conf_set(env, lli->lli_clob, conf);
3334 cl_env_nested_put(&nest, env);
3336 if (conf->coc_opc == OBJECT_CONF_SET) {
3337 struct ldlm_lock *lock = conf->coc_lock;
3340 LASSERT(ldlm_has_layout(lock));
3342 /* it can only be allowed to match after layout is
3343 * applied to inode otherwise false layout would be
3344 * seen. Applying layout should happen before dropping
3347 ldlm_lock_allow_match(lock);
3353 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3354 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3357 struct ll_sb_info *sbi = ll_i2sbi(inode);
3358 struct ptlrpc_request *req;
3359 struct mdt_body *body;
3365 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3366 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3367 lock->l_lvb_data, lock->l_lvb_len);
3369 if (lock->l_lvb_data && ldlm_is_lvb_ready(lock))
3372 /* if layout lock was granted right away, the layout is returned
3373 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3374 * blocked and then granted via completion ast, we have to fetch
3375 * layout here. Please note that we can't use the LVB buffer in
3376 * completion AST because it doesn't have a large enough buffer
3378 rc = ll_get_default_mdsize(sbi, &lmmsize);
3380 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
3381 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3386 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3392 lmmsize = body->eadatasize;
3393 if (lmmsize == 0) /* empty layout */ {
3398 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3404 lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS);
3410 memcpy(lvbdata, lmm, lmmsize);
3411 lock_res_and_lock(lock);
3412 if (lock->l_lvb_data)
3413 kvfree(lock->l_lvb_data);
3415 lock->l_lvb_data = lvbdata;
3416 lock->l_lvb_len = lmmsize;
3417 unlock_res_and_lock(lock);
3420 ptlrpc_req_finished(req);
3425 * Apply the layout to the inode. Layout lock is held and will be released
3428 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
3429 struct inode *inode, __u32 *gen, bool reconf)
3431 struct ll_inode_info *lli = ll_i2info(inode);
3432 struct ll_sb_info *sbi = ll_i2sbi(inode);
3433 struct ldlm_lock *lock;
3434 struct lustre_md md = { NULL };
3435 struct cl_object_conf conf;
3438 bool wait_layout = false;
3440 LASSERT(lustre_handle_is_used(lockh));
3442 lock = ldlm_handle2lock(lockh);
3444 LASSERT(ldlm_has_layout(lock));
3446 LDLM_DEBUG(lock, "File "DFID"(%p) being reconfigured: %d",
3447 PFID(&lli->lli_fid), inode, reconf);
3449 /* in case this is a caching lock and reinstate with new inode */
3450 md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
3452 lock_res_and_lock(lock);
3453 lvb_ready = ldlm_is_lvb_ready(lock);
3454 unlock_res_and_lock(lock);
3455 /* checking lvb_ready is racy but this is okay. The worst case is
3456 * that multi processes may configure the file on the same time.
3458 if (lvb_ready || !reconf) {
3461 /* layout_gen must be valid if layout lock is not
3462 * cancelled and stripe has already set
3464 *gen = ll_layout_version_get(lli);
3470 rc = ll_layout_fetch(inode, lock);
3474 /* for layout lock, lmm is returned in lock's lvb.
3475 * lvb_data is immutable if the lock is held so it's safe to access it
3476 * without res lock. See the description in ldlm_lock_decref_internal()
3477 * for the condition to free lvb_data of layout lock
3479 if (lock->l_lvb_data) {
3480 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3481 lock->l_lvb_data, lock->l_lvb_len);
3483 *gen = LL_LAYOUT_GEN_EMPTY;
3485 *gen = md.lsm->lsm_layout_gen;
3488 CERROR("%s: file " DFID " unpackmd error: %d\n",
3489 ll_get_fsname(inode->i_sb, NULL, 0),
3490 PFID(&lli->lli_fid), rc);
3496 /* set layout to file. Unlikely this will fail as old layout was
3499 memset(&conf, 0, sizeof(conf));
3500 conf.coc_opc = OBJECT_CONF_SET;
3501 conf.coc_inode = inode;
3502 conf.coc_lock = lock;
3503 conf.u.coc_md = &md;
3504 rc = ll_layout_conf(inode, &conf);
3507 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3509 /* refresh layout failed, need to wait */
3510 wait_layout = rc == -EBUSY;
3513 LDLM_LOCK_PUT(lock);
3514 ldlm_lock_decref(lockh, mode);
3516 /* wait for IO to complete if it's still being used. */
3518 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
3519 ll_get_fsname(inode->i_sb, NULL, 0),
3520 PFID(&lli->lli_fid), inode);
3522 memset(&conf, 0, sizeof(conf));
3523 conf.coc_opc = OBJECT_CONF_WAIT;
3524 conf.coc_inode = inode;
3525 rc = ll_layout_conf(inode, &conf);
3529 CDEBUG(D_INODE, "%s: file="DFID" waiting layout return: %d.\n",
3530 ll_get_fsname(inode->i_sb, NULL, 0),
3531 PFID(&lli->lli_fid), rc);
3537 * This function checks if there exists a LAYOUT lock on the client side,
3538 * or enqueues it if it doesn't have one in cache.
3540 * This function will not hold layout lock so it may be revoked any time after
3541 * this function returns. Any operations depend on layout should be redone
3544 * This function should be called before lov_io_init() to get an uptodate
3545 * layout version, the caller should save the version number and after IO
3546 * is finished, this function should be called again to verify that layout
3547 * is not changed during IO time.
3549 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3551 struct ll_inode_info *lli = ll_i2info(inode);
3552 struct ll_sb_info *sbi = ll_i2sbi(inode);
3553 struct md_op_data *op_data;
3554 struct lookup_intent it;
3555 struct lustre_handle lockh;
3556 enum ldlm_mode mode;
3557 struct ldlm_enqueue_info einfo = {
3558 .ei_type = LDLM_IBITS,
3560 .ei_cb_bl = ll_md_blocking_ast,
3561 .ei_cb_cp = ldlm_completion_ast,
3565 *gen = ll_layout_version_get(lli);
3566 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE)
3570 LASSERT(fid_is_sane(ll_inode2fid(inode)));
3571 LASSERT(S_ISREG(inode->i_mode));
3573 /* take layout lock mutex to enqueue layout lock exclusively. */
3574 mutex_lock(&lli->lli_layout_mutex);
3577 /* mostly layout lock is caching on the local side, so try to match
3578 * it before grabbing layout lock mutex.
3580 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
3581 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
3582 if (mode != 0) { /* hit cached lock */
3583 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3587 mutex_unlock(&lli->lli_layout_mutex);
3591 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
3592 0, 0, LUSTRE_OPC_ANY, NULL);
3593 if (IS_ERR(op_data)) {
3594 mutex_unlock(&lli->lli_layout_mutex);
3595 return PTR_ERR(op_data);
3598 /* have to enqueue one */
3599 memset(&it, 0, sizeof(it));
3600 it.it_op = IT_LAYOUT;
3601 lockh.cookie = 0ULL;
3603 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
3604 ll_get_fsname(inode->i_sb, NULL, 0),
3605 PFID(&lli->lli_fid), inode);
3607 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh,
3609 ptlrpc_req_finished(it.it_request);
3610 it.it_request = NULL;
3612 ll_finish_md_op_data(op_data);
3614 mode = it.it_lock_mode;
3615 it.it_lock_mode = 0;
3616 ll_intent_drop_lock(&it);
3619 /* set lock data in case this is a new lock */
3620 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
3621 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3625 mutex_unlock(&lli->lli_layout_mutex);
3631 * This function send a restore request to the MDT
3633 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
3635 struct hsm_user_request *hur;
3638 len = sizeof(struct hsm_user_request) +
3639 sizeof(struct hsm_user_item);
3640 hur = kzalloc(len, GFP_NOFS);
3644 hur->hur_request.hr_action = HUA_RESTORE;
3645 hur->hur_request.hr_archive_id = 0;
3646 hur->hur_request.hr_flags = 0;
3647 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
3648 sizeof(hur->hur_user_item[0].hui_fid));
3649 hur->hur_user_item[0].hui_extent.offset = offset;
3650 hur->hur_user_item[0].hui_extent.length = length;
3651 hur->hur_request.hr_itemcount = 1;
3652 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,