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) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Phil Schwan <phil@clusterfs.com>
40 * Author: Andreas Dilger <adilger@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_LLITE
44 #include "../include/lustre_dlm.h"
45 #include "../include/lustre_lite.h"
46 #include <linux/pagemap.h>
47 #include <linux/file.h>
48 #include <linux/mount.h>
49 #include "llite_internal.h"
50 #include "../include/lustre/ll_fiemap.h"
52 #include "../include/cl_object.h"
55 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
57 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
60 static enum llioc_iter
61 ll_iocontrol_call(struct inode *inode, struct file *file,
62 unsigned int cmd, unsigned long arg, int *rcp);
64 static struct ll_file_data *ll_file_data_get(void)
66 struct ll_file_data *fd;
68 fd = kmem_cache_zalloc(ll_file_data_slab, GFP_NOFS);
71 fd->fd_write_failed = false;
75 static void ll_file_data_put(struct ll_file_data *fd)
78 kmem_cache_free(ll_file_data_slab, fd);
81 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
82 struct lustre_handle *fh)
84 op_data->op_fid1 = ll_i2info(inode)->lli_fid;
85 op_data->op_attr.ia_mode = inode->i_mode;
86 op_data->op_attr.ia_atime = inode->i_atime;
87 op_data->op_attr.ia_mtime = inode->i_mtime;
88 op_data->op_attr.ia_ctime = inode->i_ctime;
89 op_data->op_attr.ia_size = i_size_read(inode);
90 op_data->op_attr_blocks = inode->i_blocks;
91 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
92 op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
94 op_data->op_handle = *fh;
96 if (ll_i2info(inode)->lli_flags & LLIF_DATA_MODIFIED)
97 op_data->op_bias |= MDS_DATA_MODIFIED;
101 * Closes the IO epoch and packs all the attributes into @op_data for
104 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
105 struct obd_client_handle *och)
107 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
108 ATTR_MTIME | ATTR_MTIME_SET |
109 ATTR_CTIME | ATTR_CTIME_SET;
111 if (!(och->och_flags & FMODE_WRITE))
114 if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
115 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
117 ll_ioepoch_close(inode, op_data, &och, 0);
120 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
121 ll_prep_md_op_data(op_data, inode, NULL, NULL,
122 0, 0, LUSTRE_OPC_ANY, NULL);
125 static int ll_close_inode_openhandle(struct obd_export *md_exp,
127 struct obd_client_handle *och,
128 const __u64 *data_version)
130 struct obd_export *exp = ll_i2mdexp(inode);
131 struct md_op_data *op_data;
132 struct ptlrpc_request *req = NULL;
133 struct obd_device *obd = class_exp2obd(exp);
139 * XXX: in case of LMV, is this correct to access
142 CERROR("Invalid MDC connection handle %#llx\n",
143 ll_i2mdexp(inode)->exp_handle.h_cookie);
148 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
150 /* XXX We leak openhandle and request here. */
155 ll_prepare_close(inode, op_data, och);
157 /* Pass in data_version implies release. */
158 op_data->op_bias |= MDS_HSM_RELEASE;
159 op_data->op_data_version = *data_version;
160 op_data->op_lease_handle = och->och_lease_handle;
161 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
163 epoch_close = op_data->op_flags & MF_EPOCH_CLOSE;
164 rc = md_close(md_exp, op_data, och->och_mod, &req);
166 /* This close must have the epoch closed. */
167 LASSERT(epoch_close);
168 /* MDS has instructed us to obtain Size-on-MDS attribute from
169 * OSTs and send setattr to back to MDS.
171 rc = ll_som_update(inode, op_data);
173 CERROR("%s: inode "DFID" mdc Size-on-MDS update failed: rc = %d\n",
174 ll_i2mdexp(inode)->exp_obd->obd_name,
175 PFID(ll_inode2fid(inode)), rc);
179 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
180 ll_i2mdexp(inode)->exp_obd->obd_name,
181 PFID(ll_inode2fid(inode)), rc);
184 /* DATA_MODIFIED flag was successfully sent on close, cancel data
187 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
188 struct ll_inode_info *lli = ll_i2info(inode);
190 spin_lock(&lli->lli_lock);
191 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
192 spin_unlock(&lli->lli_lock);
196 rc = ll_objects_destroy(req, inode);
198 CERROR("inode %lu ll_objects destroy: rc = %d\n",
201 if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) {
202 struct mdt_body *body;
204 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
205 if (!(body->valid & OBD_MD_FLRELEASED))
209 ll_finish_md_op_data(op_data);
212 if (exp_connect_som(exp) && !epoch_close &&
213 S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
214 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
216 md_clear_open_replay_data(md_exp, och);
217 /* Free @och if it is not waiting for DONE_WRITING. */
218 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
221 if (req) /* This is close request */
222 ptlrpc_req_finished(req);
226 int ll_md_real_close(struct inode *inode, fmode_t fmode)
228 struct ll_inode_info *lli = ll_i2info(inode);
229 struct obd_client_handle **och_p;
230 struct obd_client_handle *och;
234 if (fmode & FMODE_WRITE) {
235 och_p = &lli->lli_mds_write_och;
236 och_usecount = &lli->lli_open_fd_write_count;
237 } else if (fmode & FMODE_EXEC) {
238 och_p = &lli->lli_mds_exec_och;
239 och_usecount = &lli->lli_open_fd_exec_count;
241 LASSERT(fmode & FMODE_READ);
242 och_p = &lli->lli_mds_read_och;
243 och_usecount = &lli->lli_open_fd_read_count;
246 mutex_lock(&lli->lli_och_mutex);
247 if (*och_usecount > 0) {
248 /* There are still users of this handle, so skip
251 mutex_unlock(&lli->lli_och_mutex);
257 mutex_unlock(&lli->lli_och_mutex);
260 /* There might be a race and this handle may already
263 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
270 static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
273 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
274 struct ll_inode_info *lli = ll_i2info(inode);
276 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
277 struct lustre_handle lockh;
278 ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_OPEN} };
281 /* clear group lock, if present */
282 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
283 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
285 if (fd->fd_lease_och) {
288 /* Usually the lease is not released when the
289 * application crashed, we need to release here.
291 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
292 CDEBUG(rc ? D_ERROR : D_INODE,
293 "Clean up lease " DFID " %d/%d\n",
294 PFID(&lli->lli_fid), rc, lease_broken);
296 fd->fd_lease_och = NULL;
300 rc = ll_close_inode_openhandle(md_exp, inode, fd->fd_och, NULL);
305 /* Let's see if we have good enough OPEN lock on the file and if
306 * we can skip talking to MDS
309 mutex_lock(&lli->lli_och_mutex);
310 if (fd->fd_omode & FMODE_WRITE) {
312 LASSERT(lli->lli_open_fd_write_count);
313 lli->lli_open_fd_write_count--;
314 } else if (fd->fd_omode & FMODE_EXEC) {
316 LASSERT(lli->lli_open_fd_exec_count);
317 lli->lli_open_fd_exec_count--;
320 LASSERT(lli->lli_open_fd_read_count);
321 lli->lli_open_fd_read_count--;
323 mutex_unlock(&lli->lli_och_mutex);
325 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
326 LDLM_IBITS, &policy, lockmode, &lockh))
327 rc = ll_md_real_close(inode, fd->fd_omode);
330 LUSTRE_FPRIVATE(file) = NULL;
331 ll_file_data_put(fd);
336 /* While this returns an error code, fput() the caller does not, so we need
337 * to make every effort to clean up all of our state here. Also, applications
338 * rarely check close errors and even if an error is returned they will not
339 * re-try the close call.
341 int ll_file_release(struct inode *inode, struct file *file)
343 struct ll_file_data *fd;
344 struct ll_sb_info *sbi = ll_i2sbi(inode);
345 struct ll_inode_info *lli = ll_i2info(inode);
348 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
349 PFID(ll_inode2fid(inode)), inode);
351 #ifdef CONFIG_FS_POSIX_ACL
352 if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
353 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
355 if (unlikely(fd->fd_flags & LL_FILE_RMTACL)) {
356 fd->fd_flags &= ~LL_FILE_RMTACL;
357 rct_del(&sbi->ll_rct, current_pid());
358 et_search_free(&sbi->ll_et, current_pid());
363 if (!is_root_inode(inode))
364 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
365 fd = LUSTRE_FPRIVATE(file);
368 /* The last ref on @file, maybe not be the owner pid of statahead.
369 * Different processes can open the same dir, "ll_opendir_key" means:
370 * it is me that should stop the statahead thread.
372 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd &&
373 lli->lli_opendir_pid != 0)
374 ll_stop_statahead(inode, lli->lli_opendir_key);
376 if (is_root_inode(inode)) {
377 LUSTRE_FPRIVATE(file) = NULL;
378 ll_file_data_put(fd);
382 if (!S_ISDIR(inode->i_mode)) {
383 lov_read_and_clear_async_rc(lli->lli_clob);
384 lli->lli_async_rc = 0;
387 rc = ll_md_close(sbi->ll_md_exp, inode, file);
389 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
390 libcfs_debug_dumplog();
395 static int ll_intent_file_open(struct dentry *dentry, void *lmm,
396 int lmmsize, struct lookup_intent *itp)
398 struct inode *inode = d_inode(dentry);
399 struct ll_sb_info *sbi = ll_i2sbi(inode);
400 struct dentry *parent = dentry->d_parent;
401 const char *name = dentry->d_name.name;
402 const int len = dentry->d_name.len;
403 struct md_op_data *op_data;
404 struct ptlrpc_request *req;
405 __u32 opc = LUSTRE_OPC_ANY;
408 /* Usually we come here only for NFSD, and we want open lock. */
409 /* We can also get here if there was cached open handle in revalidate_it
410 * but it disappeared while we were getting from there to ll_file_open.
411 * But this means this file was closed and immediately opened which
412 * makes a good candidate for using OPEN lock
414 /* If lmmsize & lmm are not 0, we are just setting stripe info
415 * parameters. No need for the open lock
417 if (!lmm && lmmsize == 0) {
418 itp->it_flags |= MDS_OPEN_LOCK;
419 if (itp->it_flags & FMODE_WRITE)
420 opc = LUSTRE_OPC_CREATE;
423 op_data = ll_prep_md_op_data(NULL, d_inode(parent),
427 return PTR_ERR(op_data);
429 itp->it_flags |= MDS_OPEN_BY_FID;
430 rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
431 0 /*unused */, &req, ll_md_blocking_ast, 0);
432 ll_finish_md_op_data(op_data);
434 /* reason for keep own exit path - don`t flood log
435 * with messages with -ESTALE errors.
437 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
438 it_open_error(DISP_OPEN_OPEN, itp))
440 ll_release_openhandle(inode, itp);
444 if (it_disposition(itp, DISP_LOOKUP_NEG)) {
449 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
450 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
451 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
455 rc = ll_prep_inode(&inode, req, NULL, itp);
456 if (!rc && itp->d.lustre.it_lock_mode)
457 ll_set_lock_data(sbi->ll_md_exp, inode, itp, NULL);
460 ptlrpc_req_finished(req);
461 ll_intent_drop_lock(itp);
467 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
468 * not believe attributes if a few ioepoch holders exist. Attributes for
469 * previous ioepoch if new one is opened are also skipped by MDS.
471 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
473 if (ioepoch && lli->lli_ioepoch != ioepoch) {
474 lli->lli_ioepoch = ioepoch;
475 CDEBUG(D_INODE, "Epoch %llu opened on "DFID"\n",
476 ioepoch, PFID(&lli->lli_fid));
480 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
481 struct obd_client_handle *och)
483 struct ptlrpc_request *req = it->d.lustre.it_data;
484 struct mdt_body *body;
486 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
487 och->och_fh = body->handle;
488 och->och_fid = body->fid1;
489 och->och_lease_handle.cookie = it->d.lustre.it_lock_handle;
490 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
491 och->och_flags = it->it_flags;
493 return md_set_open_replay_data(md_exp, och, it);
496 static int ll_local_open(struct file *file, struct lookup_intent *it,
497 struct ll_file_data *fd, struct obd_client_handle *och)
499 struct inode *inode = file_inode(file);
500 struct ll_inode_info *lli = ll_i2info(inode);
502 LASSERT(!LUSTRE_FPRIVATE(file));
507 struct ptlrpc_request *req = it->d.lustre.it_data;
508 struct mdt_body *body;
511 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
515 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
516 ll_ioepoch_open(lli, body->ioepoch);
519 LUSTRE_FPRIVATE(file) = fd;
520 ll_readahead_init(inode, &fd->fd_ras);
521 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
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->d.lustre.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->d.lustre.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->d.lustre.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->d.lustre.it_data);
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.d.lustre.it_lock_mode == 0 ||
869 it.d.lustre.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.d.lustre.it_lock_mode,
873 it.d.lustre.it_lock_bits);
878 ll_intent_release(&it);
882 /* Cancel open lock */
883 if (it.d.lustre.it_lock_mode != 0) {
884 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
885 it.d.lustre.it_lock_mode);
886 it.d.lustre.it_lock_mode = 0;
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 result = cl_io_loop(env, io);
1182 if (args->via_io_subtype == IO_NORMAL)
1183 up_read(&lli->lli_trunc_sem);
1184 if (write_mutex_locked)
1185 mutex_unlock(&lli->lli_write_mutex);
1187 /* cl_io_rw_init() handled IO */
1188 result = io->ci_result;
1191 if (io->ci_nob > 0) {
1192 result = io->ci_nob;
1193 *ppos = io->u.ci_wr.wr.crw_pos;
1197 cl_io_fini(env, io);
1198 /* If any bit been read/written (result != 0), we just return
1199 * short read/write instead of restart io.
1201 if ((result == 0 || result == -ENODATA) && io->ci_need_restart) {
1202 CDEBUG(D_VFSTRACE, "Restart %s on %pD from %lld, count:%zd\n",
1203 iot == CIT_READ ? "read" : "write",
1204 file, *ppos, count);
1205 LASSERTF(io->ci_nob == 0, "%zd\n", io->ci_nob);
1209 if (iot == CIT_READ) {
1211 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1212 LPROC_LL_READ_BYTES, result);
1213 } else if (iot == CIT_WRITE) {
1215 ll_stats_ops_tally(ll_i2sbi(file_inode(file)),
1216 LPROC_LL_WRITE_BYTES, result);
1217 fd->fd_write_failed = false;
1218 } else if (result != -ERESTARTSYS) {
1219 fd->fd_write_failed = true;
1222 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1227 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1230 struct vvp_io_args *args;
1234 env = cl_env_get(&refcheck);
1236 return PTR_ERR(env);
1238 args = ll_env_args(env, IO_NORMAL);
1239 args->u.normal.via_iter = to;
1240 args->u.normal.via_iocb = iocb;
1242 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1243 &iocb->ki_pos, iov_iter_count(to));
1244 cl_env_put(env, &refcheck);
1249 * Write to a file (through the page cache).
1251 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1254 struct vvp_io_args *args;
1258 env = cl_env_get(&refcheck);
1260 return PTR_ERR(env);
1262 args = ll_env_args(env, IO_NORMAL);
1263 args->u.normal.via_iter = from;
1264 args->u.normal.via_iocb = iocb;
1266 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1267 &iocb->ki_pos, iov_iter_count(from));
1268 cl_env_put(env, &refcheck);
1273 * Send file content (through pagecache) somewhere with helper
1275 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1276 struct pipe_inode_info *pipe, size_t count,
1280 struct vvp_io_args *args;
1284 env = cl_env_get(&refcheck);
1286 return PTR_ERR(env);
1288 args = ll_env_args(env, IO_SPLICE);
1289 args->u.splice.via_pipe = pipe;
1290 args->u.splice.via_flags = flags;
1292 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1293 cl_env_put(env, &refcheck);
1297 static int ll_lov_recreate(struct inode *inode, struct ost_id *oi, u32 ost_idx)
1299 struct obd_export *exp = ll_i2dtexp(inode);
1300 struct obd_trans_info oti = { 0 };
1301 struct obdo *oa = NULL;
1304 struct lov_stripe_md *lsm = NULL, *lsm2;
1306 oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS);
1310 lsm = ccc_inode_lsm_get(inode);
1311 if (!lsm_has_objects(lsm)) {
1316 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1317 (lsm->lsm_stripe_count));
1319 lsm2 = libcfs_kvzalloc(lsm_size, GFP_NOFS);
1326 oa->o_nlink = ost_idx;
1327 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1328 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1329 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1330 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1331 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
1332 memcpy(lsm2, lsm, lsm_size);
1333 ll_inode_size_lock(inode);
1334 rc = obd_create(NULL, exp, oa, &lsm2, &oti);
1335 ll_inode_size_unlock(inode);
1340 ccc_inode_lsm_put(inode, lsm);
1341 kmem_cache_free(obdo_cachep, oa);
1345 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1347 struct ll_recreate_obj ucreat;
1350 if (!capable(CFS_CAP_SYS_ADMIN))
1353 if (copy_from_user(&ucreat, (struct ll_recreate_obj __user *)arg,
1357 ostid_set_seq_mdt0(&oi);
1358 ostid_set_id(&oi, ucreat.lrc_id);
1359 return ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx);
1362 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1368 if (!capable(CFS_CAP_SYS_ADMIN))
1371 if (copy_from_user(&fid, (struct lu_fid __user *)arg, sizeof(fid)))
1374 fid_to_ostid(&fid, &oi);
1375 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1376 return ll_lov_recreate(inode, &oi, ost_idx);
1379 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1380 __u64 flags, struct lov_user_md *lum,
1383 struct lov_stripe_md *lsm = NULL;
1384 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1387 lsm = ccc_inode_lsm_get(inode);
1389 ccc_inode_lsm_put(inode, lsm);
1390 CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n",
1391 PFID(ll_inode2fid(inode)));
1396 ll_inode_size_lock(inode);
1397 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
1400 rc = oit.d.lustre.it_status;
1404 ll_release_openhandle(inode, &oit);
1407 ll_inode_size_unlock(inode);
1408 ll_intent_release(&oit);
1409 ccc_inode_lsm_put(inode, lsm);
1413 ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
1417 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1418 struct lov_mds_md **lmmp, int *lmm_size,
1419 struct ptlrpc_request **request)
1421 struct ll_sb_info *sbi = ll_i2sbi(inode);
1422 struct mdt_body *body;
1423 struct lov_mds_md *lmm = NULL;
1424 struct ptlrpc_request *req = NULL;
1425 struct md_op_data *op_data;
1428 rc = ll_get_default_mdsize(sbi, &lmmsize);
1432 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1433 strlen(filename), lmmsize,
1434 LUSTRE_OPC_ANY, NULL);
1435 if (IS_ERR(op_data))
1436 return PTR_ERR(op_data);
1438 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1439 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1440 ll_finish_md_op_data(op_data);
1442 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
1447 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1449 lmmsize = body->eadatasize;
1451 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1457 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1459 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1460 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1466 * This is coming from the MDS, so is probably in
1467 * little endian. We convert it to host endian before
1468 * passing it to userspace.
1470 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
1473 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1474 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1477 /* if function called for directory - we should
1478 * avoid swab not existent lsm objects
1480 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1481 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1482 if (S_ISREG(body->mode))
1483 lustre_swab_lov_user_md_objects(
1484 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1486 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1487 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1488 if (S_ISREG(body->mode))
1489 lustre_swab_lov_user_md_objects(
1490 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1497 *lmm_size = lmmsize;
1502 static int ll_lov_setea(struct inode *inode, struct file *file,
1505 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1506 struct lov_user_md *lump;
1507 int lum_size = sizeof(struct lov_user_md) +
1508 sizeof(struct lov_user_ost_data);
1511 if (!capable(CFS_CAP_SYS_ADMIN))
1514 lump = libcfs_kvzalloc(lum_size, GFP_NOFS);
1518 if (copy_from_user(lump, (struct lov_user_md __user *)arg, lum_size)) {
1523 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lump,
1525 cl_lov_delay_create_clear(&file->f_flags);
1531 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1534 struct lov_user_md_v3 lumv3;
1535 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1536 struct lov_user_md_v1 __user *lumv1p = (void __user *)arg;
1537 struct lov_user_md_v3 __user *lumv3p = (void __user *)arg;
1539 __u64 flags = FMODE_WRITE;
1541 /* first try with v1 which is smaller than v3 */
1542 lum_size = sizeof(struct lov_user_md_v1);
1543 if (copy_from_user(lumv1, lumv1p, lum_size))
1546 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1547 lum_size = sizeof(struct lov_user_md_v3);
1548 if (copy_from_user(&lumv3, lumv3p, lum_size))
1552 rc = ll_lov_setstripe_ea_info(inode, file->f_path.dentry, flags, lumv1,
1554 cl_lov_delay_create_clear(&file->f_flags);
1556 struct lov_stripe_md *lsm;
1559 put_user(0, &lumv1p->lmm_stripe_count);
1561 ll_layout_refresh(inode, &gen);
1562 lsm = ccc_inode_lsm_get(inode);
1563 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1564 0, lsm, (void __user *)arg);
1565 ccc_inode_lsm_put(inode, lsm);
1570 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1572 struct lov_stripe_md *lsm;
1575 lsm = ccc_inode_lsm_get(inode);
1577 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
1578 lsm, (void __user *)arg);
1579 ccc_inode_lsm_put(inode, lsm);
1584 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1586 struct ll_inode_info *lli = ll_i2info(inode);
1587 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1588 struct ll_grouplock grouplock;
1592 CWARN("group id for group lock must not be 0\n");
1596 if (ll_file_nolock(file))
1599 spin_lock(&lli->lli_lock);
1600 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1601 CWARN("group lock already existed with gid %lu\n",
1602 fd->fd_grouplock.lg_gid);
1603 spin_unlock(&lli->lli_lock);
1606 LASSERT(!fd->fd_grouplock.lg_lock);
1607 spin_unlock(&lli->lli_lock);
1609 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
1610 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1614 spin_lock(&lli->lli_lock);
1615 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1616 spin_unlock(&lli->lli_lock);
1617 CERROR("another thread just won the race\n");
1618 cl_put_grouplock(&grouplock);
1622 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1623 fd->fd_grouplock = grouplock;
1624 spin_unlock(&lli->lli_lock);
1626 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1630 static int ll_put_grouplock(struct inode *inode, struct file *file,
1633 struct ll_inode_info *lli = ll_i2info(inode);
1634 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1635 struct ll_grouplock grouplock;
1637 spin_lock(&lli->lli_lock);
1638 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1639 spin_unlock(&lli->lli_lock);
1640 CWARN("no group lock held\n");
1643 LASSERT(fd->fd_grouplock.lg_lock);
1645 if (fd->fd_grouplock.lg_gid != arg) {
1646 CWARN("group lock %lu doesn't match current id %lu\n",
1647 arg, fd->fd_grouplock.lg_gid);
1648 spin_unlock(&lli->lli_lock);
1652 grouplock = fd->fd_grouplock;
1653 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1654 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1655 spin_unlock(&lli->lli_lock);
1657 cl_put_grouplock(&grouplock);
1658 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1663 * Close inode open handle
1665 * \param inode [in] inode in question
1666 * \param it [in,out] intent which contains open info and result
1669 * \retval <0 failure
1671 int ll_release_openhandle(struct inode *inode, struct lookup_intent *it)
1673 struct obd_client_handle *och;
1678 /* Root ? Do nothing. */
1679 if (is_root_inode(inode))
1682 /* No open handle to close? Move away */
1683 if (!it_disposition(it, DISP_OPEN_OPEN))
1686 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1688 och = kzalloc(sizeof(*och), GFP_NOFS);
1694 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1696 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1699 /* this one is in place of ll_file_open */
1700 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1701 ptlrpc_req_finished(it->d.lustre.it_data);
1702 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1708 * Get size for inode for which FIEMAP mapping is requested.
1709 * Make the FIEMAP get_info call and returns the result.
1711 static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1714 struct obd_export *exp = ll_i2dtexp(inode);
1715 struct lov_stripe_md *lsm = NULL;
1716 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1717 __u32 vallen = num_bytes;
1720 /* Checks for fiemap flags */
1721 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1722 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1726 /* Check for FIEMAP_FLAG_SYNC */
1727 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1728 rc = filemap_fdatawrite(inode->i_mapping);
1733 lsm = ccc_inode_lsm_get(inode);
1737 /* If the stripe_count > 1 and the application does not understand
1738 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1740 if (lsm->lsm_stripe_count > 1 &&
1741 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) {
1746 fm_key.oa.o_oi = lsm->lsm_oi;
1747 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1749 if (i_size_read(inode) == 0) {
1750 rc = ll_glimpse_size(inode);
1755 obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
1756 obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
1757 /* If filesize is 0, then there would be no objects for mapping */
1758 if (fm_key.oa.o_size == 0) {
1759 fiemap->fm_mapped_extents = 0;
1764 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1766 rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
1769 CERROR("obd_get_info failed: rc = %d\n", rc);
1772 ccc_inode_lsm_put(inode, lsm);
1776 int ll_fid2path(struct inode *inode, void __user *arg)
1778 struct obd_export *exp = ll_i2mdexp(inode);
1779 const struct getinfo_fid2path __user *gfin = arg;
1780 struct getinfo_fid2path *gfout;
1785 if (!capable(CFS_CAP_DAC_READ_SEARCH) &&
1786 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1789 /* Only need to get the buflen */
1790 if (get_user(pathlen, &gfin->gf_pathlen))
1793 if (pathlen > PATH_MAX)
1796 outsize = sizeof(*gfout) + pathlen;
1798 gfout = kzalloc(outsize, GFP_NOFS);
1802 if (copy_from_user(gfout, arg, sizeof(*gfout))) {
1807 /* Call mdc_iocontrol */
1808 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1812 if (copy_to_user(arg, gfout, outsize))
1820 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1822 struct ll_user_fiemap *fiemap_s;
1823 size_t num_bytes, ret_bytes;
1824 unsigned int extent_count;
1827 /* Get the extent count so we can calculate the size of
1828 * required fiemap buffer
1830 if (get_user(extent_count,
1831 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1835 (SIZE_MAX - sizeof(*fiemap_s)) / sizeof(struct ll_fiemap_extent))
1837 num_bytes = sizeof(*fiemap_s) + (extent_count *
1838 sizeof(struct ll_fiemap_extent));
1840 fiemap_s = libcfs_kvzalloc(num_bytes, GFP_NOFS);
1844 /* get the fiemap value */
1845 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1846 sizeof(*fiemap_s))) {
1851 /* If fm_extent_count is non-zero, read the first extent since
1852 * it is used to calculate end_offset and device from previous
1856 if (copy_from_user(&fiemap_s->fm_extents[0],
1857 (char __user *)arg + sizeof(*fiemap_s),
1858 sizeof(struct ll_fiemap_extent))) {
1864 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1868 ret_bytes = sizeof(struct ll_user_fiemap);
1870 if (extent_count != 0)
1871 ret_bytes += (fiemap_s->fm_mapped_extents *
1872 sizeof(struct ll_fiemap_extent));
1874 if (copy_to_user((void __user *)arg, fiemap_s, ret_bytes))
1883 * Read the data_version for inode.
1885 * This value is computed using stripe object version on OST.
1886 * Version is computed using server side locking.
1888 * @param sync if do sync on the OST side;
1890 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1891 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1893 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
1895 struct lov_stripe_md *lsm = NULL;
1896 struct ll_sb_info *sbi = ll_i2sbi(inode);
1897 struct obdo *obdo = NULL;
1900 /* If no stripe, we consider version is 0. */
1901 lsm = ccc_inode_lsm_get(inode);
1902 if (!lsm_has_objects(lsm)) {
1904 CDEBUG(D_INODE, "No object for inode\n");
1909 obdo = kzalloc(sizeof(*obdo), GFP_NOFS);
1915 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, obdo, 0, flags);
1917 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
1920 *data_version = obdo->o_data_version;
1925 ccc_inode_lsm_put(inode, lsm);
1930 * Trigger a HSM release request for the provided inode.
1932 int ll_hsm_release(struct inode *inode)
1934 struct cl_env_nest nest;
1936 struct obd_client_handle *och = NULL;
1937 __u64 data_version = 0;
1940 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
1941 ll_get_fsname(inode->i_sb, NULL, 0),
1942 PFID(&ll_i2info(inode)->lli_fid));
1944 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
1950 /* Grab latest data_version and [am]time values */
1951 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
1955 env = cl_env_nested_get(&nest);
1961 ll_merge_attr(env, inode);
1962 cl_env_nested_put(&nest, env);
1964 /* Release the file.
1965 * NB: lease lock handle is released in mdc_hsm_release_pack() because
1966 * we still need it to pack l_remote_handle to MDT.
1968 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
1973 if (och && !IS_ERR(och)) /* close the file */
1974 ll_lease_close(och, inode, NULL);
1979 struct ll_swap_stack {
1980 struct iattr ia1, ia2;
1982 struct inode *inode1, *inode2;
1983 bool check_dv1, check_dv2;
1986 static int ll_swap_layouts(struct file *file1, struct file *file2,
1987 struct lustre_swap_layouts *lsl)
1989 struct mdc_swap_layouts msl;
1990 struct md_op_data *op_data;
1993 struct ll_swap_stack *llss = NULL;
1996 llss = kzalloc(sizeof(*llss), GFP_NOFS);
2000 llss->inode1 = file_inode(file1);
2001 llss->inode2 = file_inode(file2);
2003 if (!S_ISREG(llss->inode2->i_mode)) {
2008 if (inode_permission(llss->inode1, MAY_WRITE) ||
2009 inode_permission(llss->inode2, MAY_WRITE)) {
2014 if (llss->inode2->i_sb != llss->inode1->i_sb) {
2019 /* we use 2 bool because it is easier to swap than 2 bits */
2020 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2021 llss->check_dv1 = true;
2023 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2024 llss->check_dv2 = true;
2026 /* we cannot use lsl->sl_dvX directly because we may swap them */
2027 llss->dv1 = lsl->sl_dv1;
2028 llss->dv2 = lsl->sl_dv2;
2030 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2031 if (rc == 0) /* same file, done! */ {
2036 if (rc < 0) { /* sequentialize it */
2037 swap(llss->inode1, llss->inode2);
2039 swap(llss->dv1, llss->dv2);
2040 swap(llss->check_dv1, llss->check_dv2);
2044 if (gid != 0) { /* application asks to flush dirty cache */
2045 rc = ll_get_grouplock(llss->inode1, file1, gid);
2049 rc = ll_get_grouplock(llss->inode2, file2, gid);
2051 ll_put_grouplock(llss->inode1, file1, gid);
2056 /* to be able to restore mtime and atime after swap
2057 * we need to first save them
2060 (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) {
2061 llss->ia1.ia_mtime = llss->inode1->i_mtime;
2062 llss->ia1.ia_atime = llss->inode1->i_atime;
2063 llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME;
2064 llss->ia2.ia_mtime = llss->inode2->i_mtime;
2065 llss->ia2.ia_atime = llss->inode2->i_atime;
2066 llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME;
2069 /* ultimate check, before swapping the layouts we check if
2070 * dataversion has changed (if requested)
2072 if (llss->check_dv1) {
2073 rc = ll_data_version(llss->inode1, &dv, 0);
2076 if (dv != llss->dv1) {
2082 if (llss->check_dv2) {
2083 rc = ll_data_version(llss->inode2, &dv, 0);
2086 if (dv != llss->dv2) {
2092 /* struct md_op_data is used to send the swap args to the mdt
2093 * only flags is missing, so we use struct mdc_swap_layouts
2094 * through the md_op_data->op_data
2096 /* flags from user space have to be converted before they are send to
2097 * server, no flag is sent today, they are only used on the client
2101 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2102 0, LUSTRE_OPC_ANY, &msl);
2103 if (IS_ERR(op_data)) {
2104 rc = PTR_ERR(op_data);
2108 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2109 sizeof(*op_data), op_data, NULL);
2110 ll_finish_md_op_data(op_data);
2114 ll_put_grouplock(llss->inode2, file2, gid);
2115 ll_put_grouplock(llss->inode1, file1, gid);
2118 /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */
2122 /* clear useless flags */
2123 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) {
2124 llss->ia1.ia_valid &= ~ATTR_MTIME;
2125 llss->ia2.ia_valid &= ~ATTR_MTIME;
2128 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) {
2129 llss->ia1.ia_valid &= ~ATTR_ATIME;
2130 llss->ia2.ia_valid &= ~ATTR_ATIME;
2133 /* update time if requested */
2135 if (llss->ia2.ia_valid != 0) {
2136 inode_lock(llss->inode1);
2137 rc = ll_setattr(file1->f_path.dentry, &llss->ia2);
2138 inode_unlock(llss->inode1);
2141 if (llss->ia1.ia_valid != 0) {
2144 inode_lock(llss->inode2);
2145 rc1 = ll_setattr(file2->f_path.dentry, &llss->ia1);
2146 inode_unlock(llss->inode2);
2157 static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2159 struct md_op_data *op_data;
2162 /* Detect out-of range masks */
2163 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2166 /* Non-root users are forbidden to set or clear flags which are
2167 * NOT defined in HSM_USER_MASK.
2169 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2170 !capable(CFS_CAP_SYS_ADMIN))
2173 /* Detect out-of range archive id */
2174 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2175 (hss->hss_archive_id > LL_HSM_MAX_ARCHIVE))
2178 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2179 LUSTRE_OPC_ANY, hss);
2180 if (IS_ERR(op_data))
2181 return PTR_ERR(op_data);
2183 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2184 sizeof(*op_data), op_data, NULL);
2186 ll_finish_md_op_data(op_data);
2191 static int ll_hsm_import(struct inode *inode, struct file *file,
2192 struct hsm_user_import *hui)
2194 struct hsm_state_set *hss = NULL;
2195 struct iattr *attr = NULL;
2198 if (!S_ISREG(inode->i_mode))
2202 hss = kzalloc(sizeof(*hss), GFP_NOFS);
2206 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2207 hss->hss_archive_id = hui->hui_archive_id;
2208 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2209 rc = ll_hsm_state_set(inode, hss);
2213 attr = kzalloc(sizeof(*attr), GFP_NOFS);
2219 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2220 attr->ia_mode |= S_IFREG;
2221 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2222 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2223 attr->ia_size = hui->hui_size;
2224 attr->ia_mtime.tv_sec = hui->hui_mtime;
2225 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2226 attr->ia_atime.tv_sec = hui->hui_atime;
2227 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2229 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2230 ATTR_UID | ATTR_GID |
2231 ATTR_MTIME | ATTR_MTIME_SET |
2232 ATTR_ATIME | ATTR_ATIME_SET;
2236 rc = ll_setattr_raw(file->f_path.dentry, attr, true);
2240 inode_unlock(inode);
2249 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2251 struct inode *inode = file_inode(file);
2252 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2255 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),cmd=%x\n",
2256 PFID(ll_inode2fid(inode)), inode, cmd);
2257 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2259 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2260 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2264 case LL_IOC_GETFLAGS:
2265 /* Get the current value of the file flags */
2266 return put_user(fd->fd_flags, (int __user *)arg);
2267 case LL_IOC_SETFLAGS:
2268 case LL_IOC_CLRFLAGS:
2269 /* Set or clear specific file flags */
2270 /* XXX This probably needs checks to ensure the flags are
2271 * not abused, and to handle any flag side effects.
2273 if (get_user(flags, (int __user *)arg))
2276 if (cmd == LL_IOC_SETFLAGS) {
2277 if ((flags & LL_FILE_IGNORE_LOCK) &&
2278 !(file->f_flags & O_DIRECT)) {
2279 CERROR("%s: unable to disable locking on non-O_DIRECT file\n",
2284 fd->fd_flags |= flags;
2286 fd->fd_flags &= ~flags;
2289 case LL_IOC_LOV_SETSTRIPE:
2290 return ll_lov_setstripe(inode, file, arg);
2291 case LL_IOC_LOV_SETEA:
2292 return ll_lov_setea(inode, file, arg);
2293 case LL_IOC_LOV_SWAP_LAYOUTS: {
2295 struct lustre_swap_layouts lsl;
2297 if (copy_from_user(&lsl, (char __user *)arg,
2298 sizeof(struct lustre_swap_layouts)))
2301 if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
2304 file2 = fget(lsl.sl_fd);
2309 if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
2310 rc = ll_swap_layouts(file, file2, &lsl);
2314 case LL_IOC_LOV_GETSTRIPE:
2315 return ll_lov_getstripe(inode, arg);
2316 case LL_IOC_RECREATE_OBJ:
2317 return ll_lov_recreate_obj(inode, arg);
2318 case LL_IOC_RECREATE_FID:
2319 return ll_lov_recreate_fid(inode, arg);
2320 case FSFILT_IOC_FIEMAP:
2321 return ll_ioctl_fiemap(inode, arg);
2322 case FSFILT_IOC_GETFLAGS:
2323 case FSFILT_IOC_SETFLAGS:
2324 return ll_iocontrol(inode, file, cmd, arg);
2325 case FSFILT_IOC_GETVERSION_OLD:
2326 case FSFILT_IOC_GETVERSION:
2327 return put_user(inode->i_generation, (int __user *)arg);
2328 case LL_IOC_GROUP_LOCK:
2329 return ll_get_grouplock(inode, file, arg);
2330 case LL_IOC_GROUP_UNLOCK:
2331 return ll_put_grouplock(inode, file, arg);
2332 case IOC_OBD_STATFS:
2333 return ll_obd_statfs(inode, (void __user *)arg);
2335 /* We need to special case any other ioctls we want to handle,
2336 * to send them to the MDS/OST as appropriate and to properly
2337 * network encode the arg field.
2338 case FSFILT_IOC_SETVERSION_OLD:
2339 case FSFILT_IOC_SETVERSION:
2341 case LL_IOC_FLUSHCTX:
2342 return ll_flush_ctx(inode);
2343 case LL_IOC_PATH2FID: {
2344 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
2345 sizeof(struct lu_fid)))
2350 case OBD_IOC_FID2PATH:
2351 return ll_fid2path(inode, (void __user *)arg);
2352 case LL_IOC_DATA_VERSION: {
2353 struct ioc_data_version idv;
2356 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
2359 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2360 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2361 if (rc == 0 && copy_to_user((char __user *)arg, &idv,
2368 case LL_IOC_GET_MDTIDX: {
2371 mdtidx = ll_get_mdt_idx(inode);
2375 if (put_user(mdtidx, (int __user *)arg))
2380 case OBD_IOC_GETDTNAME:
2381 case OBD_IOC_GETMDNAME:
2382 return ll_get_obd_name(inode, cmd, arg);
2383 case LL_IOC_HSM_STATE_GET: {
2384 struct md_op_data *op_data;
2385 struct hsm_user_state *hus;
2388 hus = kzalloc(sizeof(*hus), GFP_NOFS);
2392 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2393 LUSTRE_OPC_ANY, hus);
2394 if (IS_ERR(op_data)) {
2396 return PTR_ERR(op_data);
2399 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2402 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
2405 ll_finish_md_op_data(op_data);
2409 case LL_IOC_HSM_STATE_SET: {
2410 struct hsm_state_set *hss;
2413 hss = memdup_user((char __user *)arg, sizeof(*hss));
2415 return PTR_ERR(hss);
2417 rc = ll_hsm_state_set(inode, hss);
2422 case LL_IOC_HSM_ACTION: {
2423 struct md_op_data *op_data;
2424 struct hsm_current_action *hca;
2427 hca = kzalloc(sizeof(*hca), GFP_NOFS);
2431 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2432 LUSTRE_OPC_ANY, hca);
2433 if (IS_ERR(op_data)) {
2435 return PTR_ERR(op_data);
2438 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2441 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
2444 ll_finish_md_op_data(op_data);
2448 case LL_IOC_SET_LEASE: {
2449 struct ll_inode_info *lli = ll_i2info(inode);
2450 struct obd_client_handle *och = NULL;
2456 if (!(file->f_mode & FMODE_WRITE))
2461 if (!(file->f_mode & FMODE_READ))
2466 mutex_lock(&lli->lli_och_mutex);
2467 if (fd->fd_lease_och) {
2468 och = fd->fd_lease_och;
2469 fd->fd_lease_och = NULL;
2471 mutex_unlock(&lli->lli_och_mutex);
2474 mode = och->och_flags &
2475 (FMODE_READ|FMODE_WRITE);
2476 rc = ll_lease_close(och, inode, &lease_broken);
2477 if (rc == 0 && lease_broken)
2483 /* return the type of lease or error */
2484 return rc < 0 ? rc : (int)mode;
2489 CDEBUG(D_INODE, "Set lease with mode %d\n", mode);
2491 /* apply for lease */
2492 och = ll_lease_open(inode, file, mode, 0);
2494 return PTR_ERR(och);
2497 mutex_lock(&lli->lli_och_mutex);
2498 if (!fd->fd_lease_och) {
2499 fd->fd_lease_och = och;
2502 mutex_unlock(&lli->lli_och_mutex);
2504 /* impossible now that only excl is supported for now */
2505 ll_lease_close(och, inode, &lease_broken);
2510 case LL_IOC_GET_LEASE: {
2511 struct ll_inode_info *lli = ll_i2info(inode);
2512 struct ldlm_lock *lock = NULL;
2515 mutex_lock(&lli->lli_och_mutex);
2516 if (fd->fd_lease_och) {
2517 struct obd_client_handle *och = fd->fd_lease_och;
2519 lock = ldlm_handle2lock(&och->och_lease_handle);
2521 lock_res_and_lock(lock);
2522 if (!ldlm_is_cancel(lock))
2523 rc = och->och_flags &
2524 (FMODE_READ | FMODE_WRITE);
2525 unlock_res_and_lock(lock);
2526 LDLM_LOCK_PUT(lock);
2529 mutex_unlock(&lli->lli_och_mutex);
2532 case LL_IOC_HSM_IMPORT: {
2533 struct hsm_user_import *hui;
2535 hui = memdup_user((void __user *)arg, sizeof(*hui));
2537 return PTR_ERR(hui);
2539 rc = ll_hsm_import(inode, file, hui);
2547 if (ll_iocontrol_call(inode, file, cmd, arg, &err) ==
2551 return obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2552 (void __user *)arg);
2557 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2559 struct inode *inode = file_inode(file);
2560 loff_t retval, eof = 0;
2562 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2563 (origin == SEEK_CUR) ? file->f_pos : 0);
2564 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2565 PFID(ll_inode2fid(inode)), inode, retval, retval, origin);
2566 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2568 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2569 retval = ll_glimpse_size(inode);
2572 eof = i_size_read(inode);
2575 retval = generic_file_llseek_size(file, offset, origin,
2576 ll_file_maxbytes(inode), eof);
2580 static int ll_flush(struct file *file, fl_owner_t id)
2582 struct inode *inode = file_inode(file);
2583 struct ll_inode_info *lli = ll_i2info(inode);
2584 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2587 LASSERT(!S_ISDIR(inode->i_mode));
2589 /* catch async errors that were recorded back when async writeback
2590 * failed for pages in this mapping.
2592 rc = lli->lli_async_rc;
2593 lli->lli_async_rc = 0;
2594 err = lov_read_and_clear_async_rc(lli->lli_clob);
2598 /* The application has been told about write failure already.
2599 * Do not report failure again.
2601 if (fd->fd_write_failed)
2603 return rc ? -EIO : 0;
2607 * Called to make sure a portion of file has been written out.
2608 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2610 * Return how many pages have been written.
2612 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2613 enum cl_fsync_mode mode, int ignore_layout)
2615 struct cl_env_nest nest;
2618 struct cl_fsync_io *fio;
2621 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2622 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2625 env = cl_env_nested_get(&nest);
2627 return PTR_ERR(env);
2629 io = vvp_env_thread_io(env);
2630 io->ci_obj = ll_i2info(inode)->lli_clob;
2631 io->ci_ignore_layout = ignore_layout;
2633 /* initialize parameters for sync */
2634 fio = &io->u.ci_fsync;
2635 fio->fi_start = start;
2637 fio->fi_fid = ll_inode2fid(inode);
2638 fio->fi_mode = mode;
2639 fio->fi_nr_written = 0;
2641 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2642 result = cl_io_loop(env, io);
2644 result = io->ci_result;
2646 result = fio->fi_nr_written;
2647 cl_io_fini(env, io);
2648 cl_env_nested_put(&nest, env);
2653 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2655 struct inode *inode = file_inode(file);
2656 struct ll_inode_info *lli = ll_i2info(inode);
2657 struct ptlrpc_request *req;
2660 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2661 PFID(ll_inode2fid(inode)), inode);
2662 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2664 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2667 /* catch async errors that were recorded back when async writeback
2668 * failed for pages in this mapping.
2670 if (!S_ISDIR(inode->i_mode)) {
2671 err = lli->lli_async_rc;
2672 lli->lli_async_rc = 0;
2675 err = lov_read_and_clear_async_rc(lli->lli_clob);
2680 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
2684 ptlrpc_req_finished(req);
2686 if (S_ISREG(inode->i_mode)) {
2687 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2689 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2690 if (rc == 0 && err < 0)
2693 fd->fd_write_failed = true;
2695 fd->fd_write_failed = false;
2698 inode_unlock(inode);
2703 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2705 struct inode *inode = file_inode(file);
2706 struct ll_sb_info *sbi = ll_i2sbi(inode);
2707 struct ldlm_enqueue_info einfo = {
2708 .ei_type = LDLM_FLOCK,
2709 .ei_cb_cp = ldlm_flock_completion_ast,
2710 .ei_cbdata = file_lock,
2712 struct md_op_data *op_data;
2713 struct lustre_handle lockh = {0};
2714 ldlm_policy_data_t flock = { {0} };
2719 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
2720 PFID(ll_inode2fid(inode)), file_lock);
2722 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2724 if (file_lock->fl_flags & FL_FLOCK)
2725 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2726 else if (!(file_lock->fl_flags & FL_POSIX))
2729 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2730 flock.l_flock.pid = file_lock->fl_pid;
2731 flock.l_flock.start = file_lock->fl_start;
2732 flock.l_flock.end = file_lock->fl_end;
2734 /* Somewhat ugly workaround for svc lockd.
2735 * lockd installs custom fl_lmops->lm_compare_owner that checks
2736 * for the fl_owner to be the same (which it always is on local node
2737 * I guess between lockd processes) and then compares pid.
2738 * As such we assign pid to the owner field to make it all work,
2739 * conflict with normal locks is unlikely since pid space and
2740 * pointer space for current->files are not intersecting
2742 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2743 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2745 switch (file_lock->fl_type) {
2747 einfo.ei_mode = LCK_PR;
2750 /* An unlock request may or may not have any relation to
2751 * existing locks so we may not be able to pass a lock handle
2752 * via a normal ldlm_lock_cancel() request. The request may even
2753 * unlock a byte range in the middle of an existing lock. In
2754 * order to process an unlock request we need all of the same
2755 * information that is given with a normal read or write record
2756 * lock request. To avoid creating another ldlm unlock (cancel)
2757 * message we'll treat a LCK_NL flock request as an unlock.
2759 einfo.ei_mode = LCK_NL;
2762 einfo.ei_mode = LCK_PW;
2765 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n",
2766 file_lock->fl_type);
2781 flags = LDLM_FL_BLOCK_NOWAIT;
2787 flags = LDLM_FL_TEST_LOCK;
2788 /* Save the old mode so that if the mode in the lock changes we
2789 * can decrement the appropriate reader or writer refcount.
2791 file_lock->fl_type = einfo.ei_mode;
2794 CERROR("unknown fcntl lock command: %d\n", cmd);
2798 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2799 LUSTRE_OPC_ANY, NULL);
2800 if (IS_ERR(op_data))
2801 return PTR_ERR(op_data);
2803 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, start=%llu, end=%llu\n",
2804 PFID(ll_inode2fid(inode)), flock.l_flock.pid, flags,
2805 einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2807 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2808 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2810 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
2811 !(flags & LDLM_FL_TEST_LOCK))
2812 rc2 = locks_lock_file_wait(file, file_lock);
2814 if (rc2 && file_lock->fl_type != F_UNLCK) {
2815 einfo.ei_mode = LCK_NL;
2816 md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2817 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2821 ll_finish_md_op_data(op_data);
2827 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2833 * test if some locks matching bits and l_req_mode are acquired
2834 * - bits can be in different locks
2835 * - if found clear the common lock bits in *bits
2836 * - the bits not found, are kept in *bits
2838 * \param bits [IN] searched lock bits [IN]
2839 * \param l_req_mode [IN] searched lock mode
2840 * \retval boolean, true iff all bits are found
2842 int ll_have_md_lock(struct inode *inode, __u64 *bits,
2843 enum ldlm_mode l_req_mode)
2845 struct lustre_handle lockh;
2846 ldlm_policy_data_t policy;
2847 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
2848 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
2856 fid = &ll_i2info(inode)->lli_fid;
2857 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
2858 ldlm_lockname[mode]);
2860 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2861 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
2862 policy.l_inodebits.bits = *bits & (1 << i);
2863 if (policy.l_inodebits.bits == 0)
2866 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
2867 &policy, mode, &lockh)) {
2868 struct ldlm_lock *lock;
2870 lock = ldlm_handle2lock(&lockh);
2873 ~(lock->l_policy_data.l_inodebits.bits);
2874 LDLM_LOCK_PUT(lock);
2876 *bits &= ~policy.l_inodebits.bits;
2883 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
2884 struct lustre_handle *lockh, __u64 flags,
2885 enum ldlm_mode mode)
2887 ldlm_policy_data_t policy = { .l_inodebits = {bits} };
2891 fid = &ll_i2info(inode)->lli_fid;
2892 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2894 rc = md_lock_match(ll_i2mdexp(inode), flags | LDLM_FL_BLOCK_GRANTED,
2895 fid, LDLM_IBITS, &policy, mode, lockh);
2900 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
2902 /* Already unlinked. Just update nlink and return success */
2903 if (rc == -ENOENT) {
2905 /* This path cannot be hit for regular files unless in
2906 * case of obscure races, so no need to validate size.
2908 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2910 } else if (rc != 0) {
2911 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
2912 "%s: revalidate FID "DFID" error: rc = %d\n",
2913 ll_get_fsname(inode->i_sb, NULL, 0),
2914 PFID(ll_inode2fid(inode)), rc);
2920 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
2922 struct inode *inode = d_inode(dentry);
2923 struct ptlrpc_request *req = NULL;
2924 struct obd_export *exp;
2927 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%pd\n",
2928 PFID(ll_inode2fid(inode)), inode, dentry);
2930 exp = ll_i2mdexp(inode);
2932 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2933 * But under CMD case, it caused some lock issues, should be fixed
2934 * with new CMD ibits lock. See bug 12718
2936 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
2937 struct lookup_intent oit = { .it_op = IT_GETATTR };
2938 struct md_op_data *op_data;
2940 if (ibits == MDS_INODELOCK_LOOKUP)
2941 oit.it_op = IT_LOOKUP;
2943 /* Call getattr by fid, so do not provide name at all. */
2944 op_data = ll_prep_md_op_data(NULL, inode,
2946 LUSTRE_OPC_ANY, NULL);
2947 if (IS_ERR(op_data))
2948 return PTR_ERR(op_data);
2950 oit.it_create_mode |= M_CHECK_STALE;
2951 rc = md_intent_lock(exp, op_data, NULL, 0,
2952 /* we are not interested in name
2956 ll_md_blocking_ast, 0);
2957 ll_finish_md_op_data(op_data);
2958 oit.it_create_mode &= ~M_CHECK_STALE;
2960 rc = ll_inode_revalidate_fini(inode, rc);
2964 rc = ll_revalidate_it_finish(req, &oit, inode);
2966 ll_intent_release(&oit);
2970 /* Unlinked? Unhash dentry, so it is not picked up later by
2971 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2972 * here to preserve get_cwd functionality on 2.6.
2975 if (!d_inode(dentry)->i_nlink)
2976 d_lustre_invalidate(dentry, 0);
2978 ll_lookup_finish_locks(&oit, inode);
2979 } else if (!ll_have_md_lock(d_inode(dentry), &ibits, LCK_MINMODE)) {
2980 struct ll_sb_info *sbi = ll_i2sbi(d_inode(dentry));
2981 u64 valid = OBD_MD_FLGETATTR;
2982 struct md_op_data *op_data;
2985 if (S_ISREG(inode->i_mode)) {
2986 rc = ll_get_default_mdsize(sbi, &ealen);
2989 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2992 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2993 0, ealen, LUSTRE_OPC_ANY,
2995 if (IS_ERR(op_data))
2996 return PTR_ERR(op_data);
2998 op_data->op_valid = valid;
2999 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
3000 ll_finish_md_op_data(op_data);
3002 rc = ll_inode_revalidate_fini(inode, rc);
3006 rc = ll_prep_inode(&inode, req, NULL, NULL);
3009 ptlrpc_req_finished(req);
3013 static int ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3015 struct inode *inode = d_inode(dentry);
3018 rc = __ll_inode_revalidate(dentry, ibits);
3022 /* if object isn't regular file, don't validate size */
3023 if (!S_ISREG(inode->i_mode)) {
3024 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_atime;
3025 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_mtime;
3026 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_ctime;
3028 /* In case of restore, the MDT has the right size and has
3029 * already send it back without granting the layout lock,
3030 * inode is up-to-date so glimpse is useless.
3031 * Also to glimpse we need the layout, in case of a running
3032 * restore the MDT holds the layout lock so the glimpse will
3033 * block up to the end of restore (getattr will block)
3035 if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING))
3036 rc = ll_glimpse_size(inode);
3041 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3043 struct inode *inode = d_inode(de);
3044 struct ll_sb_info *sbi = ll_i2sbi(inode);
3045 struct ll_inode_info *lli = ll_i2info(inode);
3048 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3049 MDS_INODELOCK_LOOKUP);
3050 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3055 stat->dev = inode->i_sb->s_dev;
3056 if (ll_need_32bit_api(sbi))
3057 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3059 stat->ino = inode->i_ino;
3060 stat->mode = inode->i_mode;
3061 stat->nlink = inode->i_nlink;
3062 stat->uid = inode->i_uid;
3063 stat->gid = inode->i_gid;
3064 stat->rdev = inode->i_rdev;
3065 stat->atime = inode->i_atime;
3066 stat->mtime = inode->i_mtime;
3067 stat->ctime = inode->i_ctime;
3068 stat->blksize = 1 << inode->i_blkbits;
3070 stat->size = i_size_read(inode);
3071 stat->blocks = inode->i_blocks;
3076 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3077 __u64 start, __u64 len)
3081 struct ll_user_fiemap *fiemap;
3082 unsigned int extent_count = fieinfo->fi_extents_max;
3084 num_bytes = sizeof(*fiemap) + (extent_count *
3085 sizeof(struct ll_fiemap_extent));
3086 fiemap = libcfs_kvzalloc(num_bytes, GFP_NOFS);
3091 fiemap->fm_flags = fieinfo->fi_flags;
3092 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3093 fiemap->fm_start = start;
3094 fiemap->fm_length = len;
3095 if (extent_count > 0 &&
3096 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3097 sizeof(struct ll_fiemap_extent)) != 0) {
3102 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3104 fieinfo->fi_flags = fiemap->fm_flags;
3105 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3106 if (extent_count > 0 &&
3107 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3108 fiemap->fm_mapped_extents *
3109 sizeof(struct ll_fiemap_extent)) != 0) {
3119 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3121 struct ll_inode_info *lli = ll_i2info(inode);
3122 struct posix_acl *acl = NULL;
3124 spin_lock(&lli->lli_lock);
3125 /* VFS' acl_permission_check->check_acl will release the refcount */
3126 acl = posix_acl_dup(lli->lli_posix_acl);
3127 spin_unlock(&lli->lli_lock);
3132 int ll_inode_permission(struct inode *inode, int mask)
3136 if (mask & MAY_NOT_BLOCK)
3139 /* as root inode are NOT getting validated in lookup operation,
3140 * need to do it before permission check.
3143 if (is_root_inode(inode)) {
3144 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3145 MDS_INODELOCK_LOOKUP);
3150 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3151 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3153 if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
3154 return lustre_check_remote_perm(inode, mask);
3156 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
3157 rc = generic_permission(inode, mask);
3162 /* -o localflock - only provides locally consistent flock locks */
3163 struct file_operations ll_file_operations = {
3164 .read_iter = ll_file_read_iter,
3165 .write_iter = ll_file_write_iter,
3166 .unlocked_ioctl = ll_file_ioctl,
3167 .open = ll_file_open,
3168 .release = ll_file_release,
3169 .mmap = ll_file_mmap,
3170 .llseek = ll_file_seek,
3171 .splice_read = ll_file_splice_read,
3176 struct file_operations ll_file_operations_flock = {
3177 .read_iter = ll_file_read_iter,
3178 .write_iter = ll_file_write_iter,
3179 .unlocked_ioctl = ll_file_ioctl,
3180 .open = ll_file_open,
3181 .release = ll_file_release,
3182 .mmap = ll_file_mmap,
3183 .llseek = ll_file_seek,
3184 .splice_read = ll_file_splice_read,
3187 .flock = ll_file_flock,
3188 .lock = ll_file_flock
3191 /* These are for -o noflock - to return ENOSYS on flock calls */
3192 struct file_operations ll_file_operations_noflock = {
3193 .read_iter = ll_file_read_iter,
3194 .write_iter = ll_file_write_iter,
3195 .unlocked_ioctl = ll_file_ioctl,
3196 .open = ll_file_open,
3197 .release = ll_file_release,
3198 .mmap = ll_file_mmap,
3199 .llseek = ll_file_seek,
3200 .splice_read = ll_file_splice_read,
3203 .flock = ll_file_noflock,
3204 .lock = ll_file_noflock
3207 const struct inode_operations ll_file_inode_operations = {
3208 .setattr = ll_setattr,
3209 .getattr = ll_getattr,
3210 .permission = ll_inode_permission,
3211 .setxattr = ll_setxattr,
3212 .getxattr = ll_getxattr,
3213 .listxattr = ll_listxattr,
3214 .removexattr = ll_removexattr,
3215 .fiemap = ll_fiemap,
3216 .get_acl = ll_get_acl,
3219 /* dynamic ioctl number support routines */
3220 static struct llioc_ctl_data {
3221 struct rw_semaphore ioc_sem;
3222 struct list_head ioc_head;
3224 __RWSEM_INITIALIZER(llioc.ioc_sem),
3225 LIST_HEAD_INIT(llioc.ioc_head)
3229 struct list_head iocd_list;
3230 unsigned int iocd_size;
3231 llioc_callback_t iocd_cb;
3232 unsigned int iocd_count;
3233 unsigned int iocd_cmd[0];
3236 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3239 struct llioc_data *in_data = NULL;
3241 if (!cb || !cmd || count > LLIOC_MAX_CMD || count < 0)
3244 size = sizeof(*in_data) + count * sizeof(unsigned int);
3245 in_data = kzalloc(size, GFP_NOFS);
3249 memset(in_data, 0, sizeof(*in_data));
3250 in_data->iocd_size = size;
3251 in_data->iocd_cb = cb;
3252 in_data->iocd_count = count;
3253 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3255 down_write(&llioc.ioc_sem);
3256 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3257 up_write(&llioc.ioc_sem);
3261 EXPORT_SYMBOL(ll_iocontrol_register);
3263 void ll_iocontrol_unregister(void *magic)
3265 struct llioc_data *tmp;
3270 down_write(&llioc.ioc_sem);
3271 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3273 list_del(&tmp->iocd_list);
3274 up_write(&llioc.ioc_sem);
3280 up_write(&llioc.ioc_sem);
3282 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3284 EXPORT_SYMBOL(ll_iocontrol_unregister);
3286 static enum llioc_iter
3287 ll_iocontrol_call(struct inode *inode, struct file *file,
3288 unsigned int cmd, unsigned long arg, int *rcp)
3290 enum llioc_iter ret = LLIOC_CONT;
3291 struct llioc_data *data;
3292 int rc = -EINVAL, i;
3294 down_read(&llioc.ioc_sem);
3295 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3296 for (i = 0; i < data->iocd_count; i++) {
3297 if (cmd != data->iocd_cmd[i])
3300 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3304 if (ret == LLIOC_STOP)
3307 up_read(&llioc.ioc_sem);
3314 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3316 struct ll_inode_info *lli = ll_i2info(inode);
3317 struct cl_env_nest nest;
3324 env = cl_env_nested_get(&nest);
3326 return PTR_ERR(env);
3328 result = cl_conf_set(env, lli->lli_clob, conf);
3329 cl_env_nested_put(&nest, env);
3331 if (conf->coc_opc == OBJECT_CONF_SET) {
3332 struct ldlm_lock *lock = conf->coc_lock;
3335 LASSERT(ldlm_has_layout(lock));
3337 /* it can only be allowed to match after layout is
3338 * applied to inode otherwise false layout would be
3339 * seen. Applying layout should happen before dropping
3342 ldlm_lock_allow_match(lock);
3348 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3349 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3352 struct ll_sb_info *sbi = ll_i2sbi(inode);
3353 struct ptlrpc_request *req;
3354 struct mdt_body *body;
3360 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3361 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3362 lock->l_lvb_data, lock->l_lvb_len);
3364 if (lock->l_lvb_data && ldlm_is_lvb_ready(lock))
3367 /* if layout lock was granted right away, the layout is returned
3368 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3369 * blocked and then granted via completion ast, we have to fetch
3370 * layout here. Please note that we can't use the LVB buffer in
3371 * completion AST because it doesn't have a large enough buffer
3373 rc = ll_get_default_mdsize(sbi, &lmmsize);
3375 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
3376 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3381 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3387 lmmsize = body->eadatasize;
3388 if (lmmsize == 0) /* empty layout */ {
3393 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3399 lvbdata = libcfs_kvzalloc(lmmsize, GFP_NOFS);
3405 memcpy(lvbdata, lmm, lmmsize);
3406 lock_res_and_lock(lock);
3407 if (lock->l_lvb_data)
3408 kvfree(lock->l_lvb_data);
3410 lock->l_lvb_data = lvbdata;
3411 lock->l_lvb_len = lmmsize;
3412 unlock_res_and_lock(lock);
3415 ptlrpc_req_finished(req);
3420 * Apply the layout to the inode. Layout lock is held and will be released
3423 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
3424 struct inode *inode, __u32 *gen, bool reconf)
3426 struct ll_inode_info *lli = ll_i2info(inode);
3427 struct ll_sb_info *sbi = ll_i2sbi(inode);
3428 struct ldlm_lock *lock;
3429 struct lustre_md md = { NULL };
3430 struct cl_object_conf conf;
3433 bool wait_layout = false;
3435 LASSERT(lustre_handle_is_used(lockh));
3437 lock = ldlm_handle2lock(lockh);
3439 LASSERT(ldlm_has_layout(lock));
3441 LDLM_DEBUG(lock, "File "DFID"(%p) being reconfigured: %d",
3442 PFID(&lli->lli_fid), inode, reconf);
3444 /* in case this is a caching lock and reinstate with new inode */
3445 md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
3447 lock_res_and_lock(lock);
3448 lvb_ready = ldlm_is_lvb_ready(lock);
3449 unlock_res_and_lock(lock);
3450 /* checking lvb_ready is racy but this is okay. The worst case is
3451 * that multi processes may configure the file on the same time.
3453 if (lvb_ready || !reconf) {
3456 /* layout_gen must be valid if layout lock is not
3457 * cancelled and stripe has already set
3459 *gen = ll_layout_version_get(lli);
3465 rc = ll_layout_fetch(inode, lock);
3469 /* for layout lock, lmm is returned in lock's lvb.
3470 * lvb_data is immutable if the lock is held so it's safe to access it
3471 * without res lock. See the description in ldlm_lock_decref_internal()
3472 * for the condition to free lvb_data of layout lock
3474 if (lock->l_lvb_data) {
3475 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3476 lock->l_lvb_data, lock->l_lvb_len);
3478 *gen = LL_LAYOUT_GEN_EMPTY;
3480 *gen = md.lsm->lsm_layout_gen;
3483 CERROR("%s: file " DFID " unpackmd error: %d\n",
3484 ll_get_fsname(inode->i_sb, NULL, 0),
3485 PFID(&lli->lli_fid), rc);
3491 /* set layout to file. Unlikely this will fail as old layout was
3494 memset(&conf, 0, sizeof(conf));
3495 conf.coc_opc = OBJECT_CONF_SET;
3496 conf.coc_inode = inode;
3497 conf.coc_lock = lock;
3498 conf.u.coc_md = &md;
3499 rc = ll_layout_conf(inode, &conf);
3502 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3504 /* refresh layout failed, need to wait */
3505 wait_layout = rc == -EBUSY;
3508 LDLM_LOCK_PUT(lock);
3509 ldlm_lock_decref(lockh, mode);
3511 /* wait for IO to complete if it's still being used. */
3513 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
3514 ll_get_fsname(inode->i_sb, NULL, 0),
3515 PFID(&lli->lli_fid), inode);
3517 memset(&conf, 0, sizeof(conf));
3518 conf.coc_opc = OBJECT_CONF_WAIT;
3519 conf.coc_inode = inode;
3520 rc = ll_layout_conf(inode, &conf);
3524 CDEBUG(D_INODE, "%s: file="DFID" waiting layout return: %d.\n",
3525 ll_get_fsname(inode->i_sb, NULL, 0),
3526 PFID(&lli->lli_fid), rc);
3532 * This function checks if there exists a LAYOUT lock on the client side,
3533 * or enqueues it if it doesn't have one in cache.
3535 * This function will not hold layout lock so it may be revoked any time after
3536 * this function returns. Any operations depend on layout should be redone
3539 * This function should be called before lov_io_init() to get an uptodate
3540 * layout version, the caller should save the version number and after IO
3541 * is finished, this function should be called again to verify that layout
3542 * is not changed during IO time.
3544 int ll_layout_refresh(struct inode *inode, __u32 *gen)
3546 struct ll_inode_info *lli = ll_i2info(inode);
3547 struct ll_sb_info *sbi = ll_i2sbi(inode);
3548 struct md_op_data *op_data;
3549 struct lookup_intent it;
3550 struct lustre_handle lockh;
3551 enum ldlm_mode mode;
3552 struct ldlm_enqueue_info einfo = {
3553 .ei_type = LDLM_IBITS,
3555 .ei_cb_bl = ll_md_blocking_ast,
3556 .ei_cb_cp = ldlm_completion_ast,
3560 *gen = ll_layout_version_get(lli);
3561 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE)
3565 LASSERT(fid_is_sane(ll_inode2fid(inode)));
3566 LASSERT(S_ISREG(inode->i_mode));
3568 /* take layout lock mutex to enqueue layout lock exclusively. */
3569 mutex_lock(&lli->lli_layout_mutex);
3572 /* mostly layout lock is caching on the local side, so try to match
3573 * it before grabbing layout lock mutex.
3575 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
3576 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
3577 if (mode != 0) { /* hit cached lock */
3578 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3582 mutex_unlock(&lli->lli_layout_mutex);
3586 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
3587 0, 0, LUSTRE_OPC_ANY, NULL);
3588 if (IS_ERR(op_data)) {
3589 mutex_unlock(&lli->lli_layout_mutex);
3590 return PTR_ERR(op_data);
3593 /* have to enqueue one */
3594 memset(&it, 0, sizeof(it));
3595 it.it_op = IT_LAYOUT;
3596 lockh.cookie = 0ULL;
3598 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
3599 ll_get_fsname(inode->i_sb, NULL, 0),
3600 PFID(&lli->lli_fid), inode);
3602 rc = md_enqueue(sbi->ll_md_exp, &einfo, &it, op_data, &lockh,
3604 ptlrpc_req_finished(it.d.lustre.it_data);
3605 it.d.lustre.it_data = NULL;
3607 ll_finish_md_op_data(op_data);
3609 mode = it.d.lustre.it_lock_mode;
3610 it.d.lustre.it_lock_mode = 0;
3611 ll_intent_drop_lock(&it);
3614 /* set lock data in case this is a new lock */
3615 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
3616 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
3620 mutex_unlock(&lli->lli_layout_mutex);
3626 * This function send a restore request to the MDT
3628 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
3630 struct hsm_user_request *hur;
3633 len = sizeof(struct hsm_user_request) +
3634 sizeof(struct hsm_user_item);
3635 hur = kzalloc(len, GFP_NOFS);
3639 hur->hur_request.hr_action = HUA_RESTORE;
3640 hur->hur_request.hr_archive_id = 0;
3641 hur->hur_request.hr_flags = 0;
3642 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
3643 sizeof(hur->hur_user_item[0].hui_fid));
3644 hur->hur_user_item[0].hui_extent.offset = offset;
3645 hur->hur_user_item[0].hui_extent.length = length;
3646 hur->hur_request.hr_itemcount = 1;
3647 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,