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, 2013, 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 <lustre_dlm.h>
45 #include <lustre_lite.h>
46 #include <linux/pagemap.h>
47 #include <linux/file.h>
48 #include <linux/sched.h>
49 #include "llite_internal.h"
50 #include <lustre/ll_fiemap.h>
51 #include <lustre_ioctl.h>
53 #include "cl_object.h"
56 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
58 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
61 static enum llioc_iter
62 ll_iocontrol_call(struct inode *inode, struct file *file,
63 unsigned int cmd, unsigned long arg, int *rcp);
65 static struct ll_file_data *ll_file_data_get(void)
67 struct ll_file_data *fd;
69 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
73 fd->fd_write_failed = false;
78 static void ll_file_data_put(struct ll_file_data *fd)
81 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
84 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
85 struct lustre_handle *fh)
87 op_data->op_fid1 = ll_i2info(inode)->lli_fid;
88 op_data->op_attr.ia_mode = inode->i_mode;
89 op_data->op_attr.ia_atime = inode->i_atime;
90 op_data->op_attr.ia_mtime = inode->i_mtime;
91 op_data->op_attr.ia_ctime = inode->i_ctime;
92 op_data->op_attr.ia_size = i_size_read(inode);
93 op_data->op_attr_blocks = inode->i_blocks;
94 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags =
95 ll_inode_to_ext_flags(inode->i_flags);
96 op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
98 op_data->op_handle = *fh;
99 op_data->op_capa1 = ll_mdscapa_get(inode);
101 if (LLIF_DATA_MODIFIED & ll_i2info(inode)->lli_flags)
102 op_data->op_bias |= MDS_DATA_MODIFIED;
106 * Closes the IO epoch and packs all the attributes into @op_data for
109 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
110 struct obd_client_handle *och)
114 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
115 ATTR_MTIME | ATTR_MTIME_SET |
116 ATTR_CTIME | ATTR_CTIME_SET;
118 if (!(och->och_flags & FMODE_WRITE))
121 if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
122 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
124 ll_ioepoch_close(inode, op_data, &och, 0);
127 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
128 ll_prep_md_op_data(op_data, inode, NULL, NULL,
129 0, 0, LUSTRE_OPC_ANY, NULL);
133 static int ll_close_inode_openhandle(struct obd_export *md_exp,
135 struct obd_client_handle *och,
136 const __u64 *data_version)
138 struct obd_export *exp = ll_i2mdexp(inode);
139 struct md_op_data *op_data;
140 struct ptlrpc_request *req = NULL;
141 struct obd_device *obd = class_exp2obd(exp);
148 * XXX: in case of LMV, is this correct to access
151 CERROR("Invalid MDC connection handle "LPX64"\n",
152 ll_i2mdexp(inode)->exp_handle.h_cookie);
156 OBD_ALLOC_PTR(op_data);
158 GOTO(out, rc = -ENOMEM); // XXX We leak openhandle and request here.
160 ll_prepare_close(inode, op_data, och);
161 if (data_version != NULL) {
162 /* Pass in data_version implies release. */
163 op_data->op_bias |= MDS_HSM_RELEASE;
164 op_data->op_data_version = *data_version;
165 op_data->op_lease_handle = och->och_lease_handle;
166 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
168 epoch_close = (op_data->op_flags & MF_EPOCH_CLOSE);
169 rc = md_close(md_exp, op_data, och->och_mod, &req);
171 /* This close must have the epoch closed. */
172 LASSERT(epoch_close);
173 /* MDS has instructed us to obtain Size-on-MDS attribute from
174 * OSTs and send setattr to back to MDS. */
175 rc = ll_som_update(inode, op_data);
177 CERROR("%s: inode "DFID" mdc Size-on-MDS update"
178 " failed: rc = %d\n",
179 ll_i2mdexp(inode)->exp_obd->obd_name,
180 PFID(ll_inode2fid(inode)), rc);
184 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
185 ll_i2mdexp(inode)->exp_obd->obd_name,
186 PFID(ll_inode2fid(inode)), rc);
189 /* DATA_MODIFIED flag was successfully sent on close, cancel data
190 * modification flag. */
191 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
192 struct ll_inode_info *lli = ll_i2info(inode);
194 spin_lock(&lli->lli_lock);
195 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
196 spin_unlock(&lli->lli_lock);
200 rc = ll_objects_destroy(req, inode);
202 CERROR("%s: inode "DFID
203 " ll_objects destroy: rc = %d\n",
204 ll_i2mdexp(inode)->exp_obd->obd_name,
205 PFID(ll_inode2fid(inode)), rc);
208 if (rc == 0 && op_data->op_bias & MDS_HSM_RELEASE) {
209 struct mdt_body *body;
210 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
211 if (!(body->mbo_valid & OBD_MD_FLRELEASED))
215 ll_finish_md_op_data(op_data);
219 if (exp_connect_som(exp) && !epoch_close &&
220 S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
221 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
223 md_clear_open_replay_data(md_exp, och);
224 /* Free @och if it is not waiting for DONE_WRITING. */
225 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
228 if (req) /* This is close request */
229 ptlrpc_req_finished(req);
233 int ll_md_real_close(struct inode *inode, fmode_t fmode)
235 struct ll_inode_info *lli = ll_i2info(inode);
236 struct obd_client_handle **och_p;
237 struct obd_client_handle *och;
242 if (fmode & FMODE_WRITE) {
243 och_p = &lli->lli_mds_write_och;
244 och_usecount = &lli->lli_open_fd_write_count;
245 } else if (fmode & FMODE_EXEC) {
246 och_p = &lli->lli_mds_exec_och;
247 och_usecount = &lli->lli_open_fd_exec_count;
249 LASSERT(fmode & FMODE_READ);
250 och_p = &lli->lli_mds_read_och;
251 och_usecount = &lli->lli_open_fd_read_count;
254 mutex_lock(&lli->lli_och_mutex);
255 if (*och_usecount > 0) {
256 /* There are still users of this handle, so skip
258 mutex_unlock(&lli->lli_och_mutex);
264 mutex_unlock(&lli->lli_och_mutex);
267 /* There might be a race and this handle may already
269 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
276 static int ll_md_close(struct obd_export *md_exp, struct inode *inode,
279 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
280 struct ll_inode_info *lli = ll_i2info(inode);
284 /* clear group lock, if present */
285 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
286 ll_put_grouplock(inode, file, fd->fd_grouplock.cg_gid);
288 if (fd->fd_lease_och != NULL) {
291 /* Usually the lease is not released when the
292 * application crashed, we need to release here. */
293 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
294 CDEBUG(rc ? D_ERROR : D_INODE, "Clean up lease "DFID" %d/%d\n",
295 PFID(&lli->lli_fid), rc, lease_broken);
297 fd->fd_lease_och = NULL;
300 if (fd->fd_och != NULL) {
301 rc = ll_close_inode_openhandle(md_exp, inode, fd->fd_och, NULL);
306 /* Let's see if we have good enough OPEN lock on the file and if
307 we can skip talking to MDS */
308 if (file->f_dentry->d_inode) { /* Can this ever be false? */
310 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
311 struct lustre_handle lockh;
312 struct inode *inode = file->f_dentry->d_inode;
313 ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};
315 mutex_lock(&lli->lli_och_mutex);
316 if (fd->fd_omode & FMODE_WRITE) {
318 LASSERT(lli->lli_open_fd_write_count);
319 lli->lli_open_fd_write_count--;
320 } else if (fd->fd_omode & FMODE_EXEC) {
322 LASSERT(lli->lli_open_fd_exec_count);
323 lli->lli_open_fd_exec_count--;
326 LASSERT(lli->lli_open_fd_read_count);
327 lli->lli_open_fd_read_count--;
329 mutex_unlock(&lli->lli_och_mutex);
331 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
332 LDLM_IBITS, &policy, lockmode,
334 rc = ll_md_real_close(file->f_dentry->d_inode,
338 CERROR("Releasing a file %p with negative dentry %p. Name %s",
339 file, file->f_dentry, file->f_dentry->d_name.name);
343 LUSTRE_FPRIVATE(file) = NULL;
344 ll_file_data_put(fd);
345 ll_capa_close(inode);
350 /* While this returns an error code, fput() the caller does not, so we need
351 * to make every effort to clean up all of our state here. Also, applications
352 * rarely check close errors and even if an error is returned they will not
353 * re-try the close call.
355 int ll_file_release(struct inode *inode, struct file *file)
357 struct ll_file_data *fd;
358 struct ll_sb_info *sbi = ll_i2sbi(inode);
359 struct ll_inode_info *lli = ll_i2info(inode);
363 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
364 PFID(ll_inode2fid(inode)), inode);
366 #ifdef CONFIG_FS_POSIX_ACL
367 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
368 inode == inode->i_sb->s_root->d_inode) {
369 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
372 if (unlikely(fd->fd_flags & LL_FILE_RMTACL)) {
373 fd->fd_flags &= ~LL_FILE_RMTACL;
374 rct_del(&sbi->ll_rct, current_pid());
375 et_search_free(&sbi->ll_et, current_pid());
380 if (inode->i_sb->s_root != file->f_dentry)
381 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
382 fd = LUSTRE_FPRIVATE(file);
385 /* The last ref on @file, maybe not the the owner pid of statahead.
386 * Different processes can open the same dir, "ll_opendir_key" means:
387 * it is me that should stop the statahead thread. */
388 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd &&
389 lli->lli_opendir_pid != 0)
390 ll_stop_statahead(inode, lli->lli_opendir_key);
392 if (inode->i_sb->s_root == file->f_dentry) {
393 LUSTRE_FPRIVATE(file) = NULL;
394 ll_file_data_put(fd);
398 if (!S_ISDIR(inode->i_mode)) {
399 if (lli->lli_clob != NULL)
400 lov_read_and_clear_async_rc(lli->lli_clob);
401 lli->lli_async_rc = 0;
404 rc = ll_md_close(sbi->ll_md_exp, inode, file);
406 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
407 libcfs_debug_dumplog();
412 static int ll_intent_file_open(struct file *file, void *lmm,
413 int lmmsize, struct lookup_intent *itp)
415 struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
416 struct dentry *parent = file->f_dentry->d_parent;
417 const char *name = file->f_dentry->d_name.name;
418 const int len = file->f_dentry->d_name.len;
419 struct md_op_data *op_data;
420 struct ptlrpc_request *req = NULL;
421 __u32 opc = LUSTRE_OPC_ANY;
428 /* Usually we come here only for NFSD, and we want open lock.
429 But we can also get here with pre 2.6.15 patchless kernels, and in
430 that case that lock is also ok */
431 /* We can also get here if there was cached open handle in revalidate_it
432 * but it disappeared while we were getting from there to ll_file_open.
433 * But this means this file was closed and immediatelly opened which
434 * makes a good candidate for using OPEN lock */
435 /* If lmmsize & lmm are not 0, we are just setting stripe info
436 * parameters. No need for the open lock */
437 if (lmm == NULL && lmmsize == 0) {
438 itp->it_flags |= MDS_OPEN_LOCK;
439 if (itp->it_flags & FMODE_WRITE)
440 opc = LUSTRE_OPC_CREATE;
443 op_data = ll_prep_md_op_data(NULL, parent->d_inode,
444 file->f_dentry->d_inode, name, len,
447 RETURN(PTR_ERR(op_data));
449 op_data->op_data = lmm;
450 op_data->op_data_size = lmmsize;
452 itp->it_flags |= MDS_OPEN_BY_FID;
453 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
454 &ll_md_blocking_ast, 0);
455 ll_finish_md_op_data(op_data);
457 /* reason for keep own exit path - don`t flood log
458 * with messages with -ESTALE errors.
460 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
461 it_open_error(DISP_OPEN_OPEN, itp))
463 ll_release_openhandle(file->f_dentry, itp);
467 if (it_disposition(itp, DISP_LOOKUP_NEG))
468 GOTO(out, rc = -ENOENT);
470 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
471 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
472 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
476 rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL, itp);
477 if (!rc && itp->d.lustre.it_lock_mode)
478 ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode,
482 ptlrpc_req_finished(req);
483 ll_intent_drop_lock(itp);
489 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
490 * not believe attributes if a few ioepoch holders exist. Attributes for
491 * previous ioepoch if new one is opened are also skipped by MDS.
493 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
495 if (ioepoch && lli->lli_ioepoch != ioepoch) {
496 lli->lli_ioepoch = ioepoch;
497 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
498 ioepoch, PFID(&lli->lli_fid));
502 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
503 struct obd_client_handle *och)
505 struct ptlrpc_request *req = it->d.lustre.it_data;
506 struct mdt_body *body;
508 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
509 och->och_fh = body->mbo_handle;
510 och->och_fid = body->mbo_fid1;
511 och->och_lease_handle.cookie = it->d.lustre.it_lock_handle;
512 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
513 och->och_flags = it->it_flags;
515 return md_set_open_replay_data(md_exp, och, it);
518 static int ll_local_open(struct file *file, struct lookup_intent *it,
519 struct ll_file_data *fd, struct obd_client_handle *och)
521 struct inode *inode = file->f_dentry->d_inode;
522 struct ll_inode_info *lli = ll_i2info(inode);
525 LASSERT(!LUSTRE_FPRIVATE(file));
530 struct ptlrpc_request *req = it->d.lustre.it_data;
531 struct mdt_body *body;
534 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
538 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
539 ll_ioepoch_open(lli, body->mbo_ioepoch);
542 LUSTRE_FPRIVATE(file) = fd;
543 ll_readahead_init(inode, &fd->fd_ras);
544 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
546 /* ll_cl_context initialize */
547 rwlock_init(&fd->fd_lock);
548 INIT_LIST_HEAD(&fd->fd_lccs);
553 /* Open a file, and (for the very first open) create objects on the OSTs at
554 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
555 * creation or open until ll_lov_setstripe() ioctl is called.
557 * If we already have the stripe MD locally then we don't request it in
558 * md_open(), by passing a lmm_size = 0.
560 * It is up to the application to ensure no other processes open this file
561 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
562 * used. We might be able to avoid races of that sort by getting lli_open_sem
563 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
564 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
566 int ll_file_open(struct inode *inode, struct file *file)
568 struct ll_inode_info *lli = ll_i2info(inode);
569 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
570 .it_flags = file->f_flags };
571 struct obd_client_handle **och_p = NULL;
572 __u64 *och_usecount = NULL;
573 struct ll_file_data *fd;
574 int rc = 0, opendir_set = 0;
577 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
578 PFID(ll_inode2fid(inode)), inode, file->f_flags);
580 it = file->private_data; /* XXX: compat macro */
581 file->private_data = NULL; /* prevent ll_local_open assertion */
583 fd = ll_file_data_get();
585 GOTO(out_openerr, rc = -ENOMEM);
588 if (S_ISDIR(inode->i_mode)) {
589 spin_lock(&lli->lli_sa_lock);
590 if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL &&
591 lli->lli_opendir_pid == 0) {
592 lli->lli_opendir_key = fd;
593 lli->lli_opendir_pid = current_pid();
596 spin_unlock(&lli->lli_sa_lock);
599 if (inode->i_sb->s_root == file->f_dentry) {
600 LUSTRE_FPRIVATE(file) = fd;
604 if (!it || !it->d.lustre.it_disposition) {
605 /* Convert f_flags into access mode. We cannot use file->f_mode,
606 * because everything but O_ACCMODE mask was stripped from
608 if ((oit.it_flags + 1) & O_ACCMODE)
610 if (file->f_flags & O_TRUNC)
611 oit.it_flags |= FMODE_WRITE;
613 /* kernel only call f_op->open in dentry_open. filp_open calls
614 * dentry_open after call to open_namei that checks permissions.
615 * Only nfsd_open call dentry_open directly without checking
616 * permissions and because of that this code below is safe. */
617 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
618 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
620 /* We do not want O_EXCL here, presumably we opened the file
621 * already? XXX - NFS implications? */
622 oit.it_flags &= ~O_EXCL;
624 /* bug20584, if "it_flags" contains O_CREAT, the file will be
625 * created if necessary, then "IT_CREAT" should be set to keep
626 * consistent with it */
627 if (oit.it_flags & O_CREAT)
628 oit.it_op |= IT_CREAT;
634 /* Let's see if we have file open on MDS already. */
635 if (it->it_flags & FMODE_WRITE) {
636 och_p = &lli->lli_mds_write_och;
637 och_usecount = &lli->lli_open_fd_write_count;
638 } else if (it->it_flags & FMODE_EXEC) {
639 och_p = &lli->lli_mds_exec_och;
640 och_usecount = &lli->lli_open_fd_exec_count;
642 och_p = &lli->lli_mds_read_och;
643 och_usecount = &lli->lli_open_fd_read_count;
646 mutex_lock(&lli->lli_och_mutex);
647 if (*och_p) { /* Open handle is present */
648 if (it_disposition(it, DISP_OPEN_OPEN)) {
649 /* Well, there's extra open request that we do not need,
650 let's close it somehow. This will decref request. */
651 rc = it_open_error(DISP_OPEN_OPEN, it);
653 mutex_unlock(&lli->lli_och_mutex);
654 GOTO(out_openerr, rc);
657 ll_release_openhandle(file->f_dentry, it);
661 rc = ll_local_open(file, it, fd, NULL);
664 mutex_unlock(&lli->lli_och_mutex);
665 GOTO(out_openerr, rc);
668 LASSERT(*och_usecount == 0);
669 if (!it->d.lustre.it_disposition) {
670 /* We cannot just request lock handle now, new ELC code
671 means that one of other OPEN locks for this file
672 could be cancelled, and since blocking ast handler
673 would attempt to grab och_mutex as well, that would
674 result in a deadlock */
675 mutex_unlock(&lli->lli_och_mutex);
676 it->it_create_mode |= M_CHECK_STALE;
677 rc = ll_intent_file_open(file, NULL, 0, it);
678 it->it_create_mode &= ~M_CHECK_STALE;
680 GOTO(out_openerr, rc);
684 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
686 GOTO(out_och_free, rc = -ENOMEM);
690 /* md_intent_lock() didn't get a request ref if there was an
691 * open error, so don't do cleanup on the request here
693 /* XXX (green): Should not we bail out on any error here, not
694 * just open error? */
695 rc = it_open_error(DISP_OPEN_OPEN, it);
697 GOTO(out_och_free, rc);
699 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
700 "inode %p: disposition %x, status %d\n", inode,
701 it_disposition(it, ~0), it->d.lustre.it_status);
703 rc = ll_local_open(file, it, fd, *och_p);
705 GOTO(out_och_free, rc);
707 mutex_unlock(&lli->lli_och_mutex);
710 /* Must do this outside lli_och_mutex lock to prevent deadlock where
711 different kind of OPEN lock for this same inode gets cancelled
712 by ldlm_cancel_lru */
713 if (!S_ISREG(inode->i_mode))
714 GOTO(out_och_free, rc);
718 if (!lli->lli_has_smd &&
719 (cl_is_lov_delay_create(file->f_flags) ||
720 (file->f_mode & FMODE_WRITE) == 0)) {
721 CDEBUG(D_INODE, "object creation was delayed\n");
722 GOTO(out_och_free, rc);
724 cl_lov_delay_create_clear(&file->f_flags);
725 GOTO(out_och_free, rc);
729 if (och_p && *och_p) {
730 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
731 *och_p = NULL; /* OBD_FREE writes some magic there */
734 mutex_unlock(&lli->lli_och_mutex);
737 if (opendir_set != 0)
738 ll_stop_statahead(inode, lli->lli_opendir_key);
740 ll_file_data_put(fd);
742 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
745 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
746 ptlrpc_req_finished(it->d.lustre.it_data);
747 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
753 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
754 struct ldlm_lock_desc *desc, void *data, int flag)
757 struct lustre_handle lockh;
761 case LDLM_CB_BLOCKING:
762 ldlm_lock2handle(lock, &lockh);
763 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
765 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
769 case LDLM_CB_CANCELING:
777 * Acquire a lease and open the file.
779 static struct obd_client_handle *
780 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
783 struct lookup_intent it = { .it_op = IT_OPEN };
784 struct ll_sb_info *sbi = ll_i2sbi(inode);
785 struct md_op_data *op_data;
786 struct ptlrpc_request *req = NULL;
787 struct lustre_handle old_handle = { 0 };
788 struct obd_client_handle *och = NULL;
793 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
794 RETURN(ERR_PTR(-EINVAL));
797 struct ll_inode_info *lli = ll_i2info(inode);
798 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
799 struct obd_client_handle **och_p;
802 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
803 RETURN(ERR_PTR(-EPERM));
805 /* Get the openhandle of the file */
807 mutex_lock(&lli->lli_och_mutex);
808 if (fd->fd_lease_och != NULL) {
809 mutex_unlock(&lli->lli_och_mutex);
813 if (fd->fd_och == NULL) {
814 if (file->f_mode & FMODE_WRITE) {
815 LASSERT(lli->lli_mds_write_och != NULL);
816 och_p = &lli->lli_mds_write_och;
817 och_usecount = &lli->lli_open_fd_write_count;
819 LASSERT(lli->lli_mds_read_och != NULL);
820 och_p = &lli->lli_mds_read_och;
821 och_usecount = &lli->lli_open_fd_read_count;
823 if (*och_usecount == 1) {
830 mutex_unlock(&lli->lli_och_mutex);
831 if (rc < 0) /* more than 1 opener */
834 LASSERT(fd->fd_och != NULL);
835 old_handle = fd->fd_och->och_fh;
840 RETURN(ERR_PTR(-ENOMEM));
842 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
843 LUSTRE_OPC_ANY, NULL);
845 GOTO(out, rc = PTR_ERR(op_data));
847 /* To tell the MDT this openhandle is from the same owner */
848 op_data->op_handle = old_handle;
850 it.it_flags = fmode | open_flags;
851 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
852 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
853 &ll_md_blocking_lease_ast,
854 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
855 * it can be cancelled which may mislead applications that the lease is
857 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
858 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
859 * doesn't deal with openhandle, so normal openhandle will be leaked. */
860 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
861 ll_finish_md_op_data(op_data);
862 ptlrpc_req_finished(req);
864 GOTO(out_release_it, rc);
866 if (it_disposition(&it, DISP_LOOKUP_NEG))
867 GOTO(out_release_it, rc = -ENOENT);
869 rc = it_open_error(DISP_OPEN_OPEN, &it);
871 GOTO(out_release_it, rc);
873 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
874 ll_och_fill(sbi->ll_md_exp, &it, och);
876 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
877 GOTO(out_close, rc = -EOPNOTSUPP);
879 /* already get lease, handle lease lock */
880 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
881 if (it.d.lustre.it_lock_mode == 0 ||
882 it.d.lustre.it_lock_bits != MDS_INODELOCK_OPEN) {
883 /* open lock must return for lease */
884 CERROR(DFID "lease granted but no open lock, %d/"LPU64".\n",
885 PFID(ll_inode2fid(inode)), it.d.lustre.it_lock_mode,
886 it.d.lustre.it_lock_bits);
887 GOTO(out_close, rc = -EPROTO);
890 ll_intent_release(&it);
894 /* Cancel open lock */
895 if (it.d.lustre.it_lock_mode != 0) {
896 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
897 it.d.lustre.it_lock_mode);
898 it.d.lustre.it_lock_mode = 0;
899 och->och_lease_handle.cookie = 0ULL;
901 rc2 = ll_close_inode_openhandle(sbi->ll_md_exp, inode, och, NULL);
903 CERROR("%s: error closing file "DFID": %d\n",
904 ll_get_fsname(inode->i_sb, NULL, 0),
905 PFID(&ll_i2info(inode)->lli_fid), rc2);
906 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
908 ll_intent_release(&it);
916 * Release lease and close the file.
917 * It will check if the lease has ever broken.
919 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
922 struct ldlm_lock *lock;
923 bool cancelled = true;
927 lock = ldlm_handle2lock(&och->och_lease_handle);
929 lock_res_and_lock(lock);
930 cancelled = ldlm_is_cancel(lock);
931 unlock_res_and_lock(lock);
935 CDEBUG(D_INODE, "lease for "DFID" broken? %d\n",
936 PFID(&ll_i2info(inode)->lli_fid), cancelled);
939 ldlm_cli_cancel(&och->och_lease_handle, 0);
940 if (lease_broken != NULL)
941 *lease_broken = cancelled;
943 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
948 /* Fills the obdo with the attributes for the lsm */
949 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
950 struct obd_capa *capa, struct obdo *obdo,
951 __u64 ioepoch, int dv_flags)
953 struct ptlrpc_request_set *set;
954 struct obd_info oinfo = { { { 0 } } };
959 LASSERT(lsm != NULL);
963 oinfo.oi_oa->o_oi = lsm->lsm_oi;
964 oinfo.oi_oa->o_mode = S_IFREG;
965 oinfo.oi_oa->o_ioepoch = ioepoch;
966 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
967 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
968 OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
969 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
970 OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
971 OBD_MD_FLDATAVERSION;
972 oinfo.oi_capa = capa;
973 if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) {
974 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
975 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
976 if (dv_flags & LL_DV_WR_FLUSH)
977 oinfo.oi_oa->o_flags |= OBD_FL_FLUSH;
980 set = ptlrpc_prep_set();
982 CERROR("can't allocate ptlrpc set\n");
985 rc = obd_getattr_async(exp, &oinfo, set);
987 rc = ptlrpc_set_wait(set);
988 ptlrpc_set_destroy(set);
991 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
992 OBD_MD_FLATIME | OBD_MD_FLMTIME |
993 OBD_MD_FLCTIME | OBD_MD_FLSIZE |
994 OBD_MD_FLDATAVERSION | OBD_MD_FLFLAGS);
995 if (dv_flags & LL_DV_WR_FLUSH &&
996 !(oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
997 oinfo.oi_oa->o_flags & OBD_FL_FLUSH))
1004 * Performs the getattr on the inode and updates its fields.
1005 * If @sync != 0, perform the getattr under the server-side lock.
1007 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
1008 __u64 ioepoch, int sync)
1010 struct obd_capa *capa = ll_mdscapa_get(inode);
1011 struct lov_stripe_md *lsm;
1015 lsm = ccc_inode_lsm_get(inode);
1016 rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode),
1017 capa, obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0);
1020 struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi;
1022 obdo_refresh_inode(inode, obdo, obdo->o_valid);
1023 CDEBUG(D_INODE, "objid "DOSTID" size %llu, blocks %llu,"
1024 " blksize %lu\n", POSTID(oi), i_size_read(inode),
1025 (unsigned long long)inode->i_blocks,
1026 (unsigned long)ll_inode_blksize(inode));
1028 ccc_inode_lsm_put(inode, lsm);
1032 int ll_merge_lvb(const struct lu_env *env, struct inode *inode)
1034 struct ll_inode_info *lli = ll_i2info(inode);
1035 struct cl_object *obj = lli->lli_clob;
1036 struct cl_attr *attr = ccc_env_thread_attr(env);
1042 ll_inode_size_lock(inode);
1043 /* merge timestamps the most recently obtained from mds with
1044 timestamps obtained from osts */
1045 LTIME_S(inode->i_atime) = lli->lli_lvb.lvb_atime;
1046 LTIME_S(inode->i_mtime) = lli->lli_lvb.lvb_mtime;
1047 LTIME_S(inode->i_ctime) = lli->lli_lvb.lvb_ctime;
1048 inode_init_lvb(inode, &lvb);
1050 cl_object_attr_lock(obj);
1051 rc = cl_object_attr_get(env, obj, attr);
1052 cl_object_attr_unlock(obj);
1055 if (lvb.lvb_atime < attr->cat_atime)
1056 lvb.lvb_atime = attr->cat_atime;
1057 if (lvb.lvb_ctime < attr->cat_ctime)
1058 lvb.lvb_ctime = attr->cat_ctime;
1059 if (lvb.lvb_mtime < attr->cat_mtime)
1060 lvb.lvb_mtime = attr->cat_mtime;
1062 CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n",
1063 PFID(&lli->lli_fid), attr->cat_size);
1064 cl_isize_write_nolock(inode, attr->cat_size);
1066 inode->i_blocks = attr->cat_blocks;
1068 LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
1069 LTIME_S(inode->i_atime) = lvb.lvb_atime;
1070 LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
1072 ll_inode_size_unlock(inode);
1077 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
1080 struct obdo obdo = { 0 };
1083 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0);
1085 st->st_size = obdo.o_size;
1086 st->st_blocks = obdo.o_blocks;
1087 st->st_mtime = obdo.o_mtime;
1088 st->st_atime = obdo.o_atime;
1089 st->st_ctime = obdo.o_ctime;
1094 static bool file_is_noatime(const struct file *file)
1096 const struct vfsmount *mnt = file->f_path.mnt;
1097 const struct inode *inode = file->f_path.dentry->d_inode;
1099 /* Adapted from file_accessed() and touch_atime().*/
1100 if (file->f_flags & O_NOATIME)
1103 if (inode->i_flags & S_NOATIME)
1106 if (IS_NOATIME(inode))
1109 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1112 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1115 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1121 static void ll_io_init(struct cl_io *io, const struct file *file, int write)
1123 struct inode *inode = file->f_dentry->d_inode;
1125 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1127 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1128 io->u.ci_wr.wr_sync = file->f_flags & O_SYNC ||
1129 file->f_flags & O_DIRECT ||
1132 io->ci_obj = ll_i2info(inode)->lli_clob;
1133 io->ci_lockreq = CILR_MAYBE;
1134 if (ll_file_nolock(file)) {
1135 io->ci_lockreq = CILR_NEVER;
1136 io->ci_no_srvlock = 1;
1137 } else if (file->f_flags & O_APPEND) {
1138 io->ci_lockreq = CILR_MANDATORY;
1141 io->ci_noatime = file_is_noatime(file);
1145 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1146 struct file *file, enum cl_io_type iot,
1147 loff_t *ppos, size_t count)
1149 struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode);
1150 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1155 CDEBUG(D_VFSTRACE, "file: %s, type: %d ppos: "LPU64", count: %zd\n",
1156 file->f_dentry->d_name.name, iot, *ppos, count);
1159 io = ccc_env_thread_io(env);
1160 ll_io_init(io, file, iot == CIT_WRITE);
1162 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1163 struct vvp_io *vio = vvp_env_io(env);
1164 struct ccc_io *cio = ccc_env_io(env);
1165 int write_mutex_locked = 0;
1167 cio->cui_fd = LUSTRE_FPRIVATE(file);
1168 vio->cui_io_subtype = args->via_io_subtype;
1170 ll_cl_add(file, env, io);
1172 switch (vio->cui_io_subtype) {
1174 cio->cui_iov = args->u.normal.via_iov;
1175 cio->cui_nrsegs = args->u.normal.via_nrsegs;
1176 cio->cui_tot_nrsegs = cio->cui_nrsegs;
1177 cio->cui_iocb = args->u.normal.via_iocb;
1178 if ((iot == CIT_WRITE) &&
1179 !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1180 if (mutex_lock_interruptible(&lli->
1182 GOTO(out, result = -ERESTARTSYS);
1183 write_mutex_locked = 1;
1185 down_read(&lli->lli_trunc_sem);
1188 vio->u.splice.cui_pipe = args->u.splice.via_pipe;
1189 vio->u.splice.cui_flags = args->u.splice.via_flags;
1192 CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
1195 result = cl_io_loop(env, io);
1196 if (args->via_io_subtype == IO_NORMAL)
1197 up_read(&lli->lli_trunc_sem);
1198 if (write_mutex_locked)
1199 mutex_unlock(&lli->lli_write_mutex);
1200 ll_cl_remove(file, env);
1202 /* cl_io_rw_init() handled IO */
1203 result = io->ci_result;
1206 if (io->ci_nob > 0) {
1207 result = io->ci_nob;
1208 *ppos = io->u.ci_wr.wr.crw_pos;
1212 cl_io_fini(env, io);
1213 /* If any bit been read/written (result != 0), we just return
1214 * short read/write instead of restart io. */
1215 if ((result == 0 || result == -ENODATA) && io->ci_need_restart) {
1216 CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n",
1217 iot == CIT_READ ? "read" : "write",
1218 file->f_dentry->d_name.name, *ppos, count);
1219 LASSERTF(io->ci_nob == 0, "%zd", io->ci_nob);
1223 if (iot == CIT_READ) {
1225 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
1226 LPROC_LL_READ_BYTES, result);
1227 } else if (iot == CIT_WRITE) {
1229 ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode),
1230 LPROC_LL_WRITE_BYTES, result);
1231 fd->fd_write_failed = false;
1232 } else if (result != -ERESTARTSYS) {
1233 fd->fd_write_failed = true;
1236 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1243 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1245 static int ll_file_get_iov_count(const struct iovec *iov,
1246 unsigned long *nr_segs, size_t *count)
1251 for (seg = 0; seg < *nr_segs; seg++) {
1252 const struct iovec *iv = &iov[seg];
1255 * If any segment has a negative length, or the cumulative
1256 * length ever wraps negative then return -EINVAL.
1259 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1261 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1266 cnt -= iv->iov_len; /* This segment is no good */
1273 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1274 unsigned long nr_segs, loff_t pos)
1277 struct vvp_io_args *args;
1283 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1287 env = cl_env_get(&refcheck);
1289 RETURN(PTR_ERR(env));
1291 args = vvp_env_args(env, IO_NORMAL);
1292 args->u.normal.via_iov = (struct iovec *)iov;
1293 args->u.normal.via_nrsegs = nr_segs;
1294 args->u.normal.via_iocb = iocb;
1296 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1297 &iocb->ki_pos, count);
1298 cl_env_put(env, &refcheck);
1302 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
1306 struct iovec *local_iov;
1307 struct kiocb *kiocb;
1312 env = cl_env_get(&refcheck);
1314 RETURN(PTR_ERR(env));
1316 local_iov = &vvp_env_info(env)->vti_local_iov;
1317 kiocb = &vvp_env_info(env)->vti_kiocb;
1318 local_iov->iov_base = (void __user *)buf;
1319 local_iov->iov_len = count;
1320 init_sync_kiocb(kiocb, file);
1321 kiocb->ki_pos = *ppos;
1322 #ifdef HAVE_KIOCB_KI_LEFT
1323 kiocb->ki_left = count;
1325 kiocb->ki_nbytes = count;
1328 result = ll_file_aio_read(kiocb, local_iov, 1, kiocb->ki_pos);
1329 *ppos = kiocb->ki_pos;
1331 cl_env_put(env, &refcheck);
1336 * Write to a file (through the page cache).
1339 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1340 unsigned long nr_segs, loff_t pos)
1343 struct vvp_io_args *args;
1349 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1353 env = cl_env_get(&refcheck);
1355 RETURN(PTR_ERR(env));
1357 args = vvp_env_args(env, IO_NORMAL);
1358 args->u.normal.via_iov = (struct iovec *)iov;
1359 args->u.normal.via_nrsegs = nr_segs;
1360 args->u.normal.via_iocb = iocb;
1362 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1363 &iocb->ki_pos, count);
1364 cl_env_put(env, &refcheck);
1368 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1372 struct iovec *local_iov;
1373 struct kiocb *kiocb;
1378 env = cl_env_get(&refcheck);
1380 RETURN(PTR_ERR(env));
1382 local_iov = &vvp_env_info(env)->vti_local_iov;
1383 kiocb = &vvp_env_info(env)->vti_kiocb;
1384 local_iov->iov_base = (void __user *)buf;
1385 local_iov->iov_len = count;
1386 init_sync_kiocb(kiocb, file);
1387 kiocb->ki_pos = *ppos;
1388 #ifdef HAVE_KIOCB_KI_LEFT
1389 kiocb->ki_left = count;
1391 kiocb->ki_nbytes = count;
1394 result = ll_file_aio_write(kiocb, local_iov, 1, kiocb->ki_pos);
1395 *ppos = kiocb->ki_pos;
1397 cl_env_put(env, &refcheck);
1402 * Send file content (through pagecache) somewhere with helper
1404 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1405 struct pipe_inode_info *pipe, size_t count,
1409 struct vvp_io_args *args;
1414 env = cl_env_get(&refcheck);
1416 RETURN(PTR_ERR(env));
1418 args = vvp_env_args(env, IO_SPLICE);
1419 args->u.splice.via_pipe = pipe;
1420 args->u.splice.via_flags = flags;
1422 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1423 cl_env_put(env, &refcheck);
1427 static int ll_lov_recreate(struct inode *inode, struct ost_id *oi,
1430 struct obd_export *exp = ll_i2dtexp(inode);
1431 struct obd_trans_info oti = { 0 };
1432 struct obdo *oa = NULL;
1435 struct lov_stripe_md *lsm = NULL, *lsm2;
1442 lsm = ccc_inode_lsm_get(inode);
1443 if (!lsm_has_objects(lsm))
1444 GOTO(out, rc = -ENOENT);
1446 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1447 (lsm->lsm_stripe_count));
1449 OBD_ALLOC_LARGE(lsm2, lsm_size);
1451 GOTO(out, rc = -ENOMEM);
1454 oa->o_nlink = ost_idx;
1455 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1456 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1457 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1458 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1459 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
1460 memcpy(lsm2, lsm, lsm_size);
1461 ll_inode_size_lock(inode);
1462 rc = obd_create(NULL, exp, oa, &lsm2, &oti);
1463 ll_inode_size_unlock(inode);
1465 OBD_FREE_LARGE(lsm2, lsm_size);
1468 ccc_inode_lsm_put(inode, lsm);
1473 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1475 struct ll_recreate_obj ucreat;
1479 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1482 if (copy_from_user(&ucreat, (struct ll_recreate_obj *)arg,
1486 ostid_set_seq_mdt0(&oi);
1487 ostid_set_id(&oi, ucreat.lrc_id);
1488 RETURN(ll_lov_recreate(inode, &oi, ucreat.lrc_ost_idx));
1491 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1498 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1501 if (copy_from_user(&fid, (struct lu_fid *)arg, sizeof(fid)))
1504 fid_to_ostid(&fid, &oi);
1505 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1506 RETURN(ll_lov_recreate(inode, &oi, ost_idx));
1509 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1510 __u64 flags, struct lov_user_md *lum,
1513 struct lov_stripe_md *lsm = NULL;
1514 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1518 lsm = ccc_inode_lsm_get(inode);
1520 ccc_inode_lsm_put(inode, lsm);
1521 CDEBUG(D_IOCTL, "stripe already exists for inode "DFID"\n",
1522 PFID(ll_inode2fid(inode)));
1523 GOTO(out, rc = -EEXIST);
1526 ll_inode_size_lock(inode);
1527 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1529 GOTO(out_unlock, rc);
1530 rc = oit.d.lustre.it_status;
1532 GOTO(out_req_free, rc);
1534 ll_release_openhandle(file->f_dentry, &oit);
1537 ll_inode_size_unlock(inode);
1538 ll_intent_release(&oit);
1539 ccc_inode_lsm_put(inode, lsm);
1541 cl_lov_delay_create_clear(&file->f_flags);
1544 ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
1548 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1549 struct lov_mds_md **lmmp, int *lmm_size,
1550 struct ptlrpc_request **request)
1552 struct ll_sb_info *sbi = ll_i2sbi(inode);
1553 struct mdt_body *body;
1554 struct lov_mds_md *lmm = NULL;
1555 struct ptlrpc_request *req = NULL;
1556 struct md_op_data *op_data;
1559 rc = ll_get_default_mdsize(sbi, &lmmsize);
1563 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1564 strlen(filename), lmmsize,
1565 LUSTRE_OPC_ANY, NULL);
1566 if (IS_ERR(op_data))
1567 RETURN(PTR_ERR(op_data));
1569 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1570 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1571 ll_finish_md_op_data(op_data);
1573 CDEBUG(D_INFO, "md_getattr_name failed "
1574 "on %s: rc %d\n", filename, rc);
1578 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1579 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1581 lmmsize = body->mbo_eadatasize;
1583 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1585 GOTO(out, rc = -ENODATA);
1588 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1589 LASSERT(lmm != NULL);
1591 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1592 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1593 GOTO(out, rc = -EPROTO);
1597 * This is coming from the MDS, so is probably in
1598 * little endian. We convert it to host endian before
1599 * passing it to userspace.
1601 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
1604 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
1605 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
1608 /* if function called for directory - we should
1609 * avoid swab not existent lsm objects */
1610 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1611 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1612 if (S_ISREG(body->mbo_mode))
1613 lustre_swab_lov_user_md_objects(
1614 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1616 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1617 lustre_swab_lov_user_md_v3(
1618 (struct lov_user_md_v3 *)lmm);
1619 if (S_ISREG(body->mbo_mode))
1620 lustre_swab_lov_user_md_objects(
1621 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1628 *lmm_size = lmmsize;
1633 static int ll_lov_setea(struct inode *inode, struct file *file,
1636 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1637 struct lov_user_md *lump;
1638 int lum_size = sizeof(struct lov_user_md) +
1639 sizeof(struct lov_user_ost_data);
1643 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1646 OBD_ALLOC_LARGE(lump, lum_size);
1650 if (copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) {
1651 OBD_FREE_LARGE(lump, lum_size);
1655 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1657 OBD_FREE_LARGE(lump, lum_size);
1661 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1664 struct lov_user_md_v3 lumv3;
1665 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1666 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1667 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1669 __u64 flags = FMODE_WRITE;
1672 /* first try with v1 which is smaller than v3 */
1673 lum_size = sizeof(struct lov_user_md_v1);
1674 if (copy_from_user(lumv1, lumv1p, lum_size))
1677 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1678 lum_size = sizeof(struct lov_user_md_v3);
1679 if (copy_from_user(&lumv3, lumv3p, lum_size))
1683 rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size);
1685 struct lov_stripe_md *lsm;
1688 put_user(0, &lumv1p->lmm_stripe_count);
1690 ll_layout_refresh(inode, &gen);
1691 lsm = ccc_inode_lsm_get(inode);
1692 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1693 0, lsm, (void *)arg);
1694 ccc_inode_lsm_put(inode, lsm);
1699 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1701 struct lov_stripe_md *lsm;
1705 lsm = ccc_inode_lsm_get(inode);
1707 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0,
1709 ccc_inode_lsm_put(inode, lsm);
1714 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1716 struct ll_inode_info *lli = ll_i2info(inode);
1717 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1718 struct ccc_grouplock grouplock;
1722 if (ll_file_nolock(file))
1723 RETURN(-EOPNOTSUPP);
1725 spin_lock(&lli->lli_lock);
1726 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1727 CWARN("group lock already existed with gid %lu\n",
1728 fd->fd_grouplock.cg_gid);
1729 spin_unlock(&lli->lli_lock);
1732 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1733 spin_unlock(&lli->lli_lock);
1735 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1736 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1740 spin_lock(&lli->lli_lock);
1741 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1742 spin_unlock(&lli->lli_lock);
1743 CERROR("another thread just won the race\n");
1744 cl_put_grouplock(&grouplock);
1748 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1749 fd->fd_grouplock = grouplock;
1750 spin_unlock(&lli->lli_lock);
1752 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1756 int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1758 struct ll_inode_info *lli = ll_i2info(inode);
1759 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1760 struct ccc_grouplock grouplock;
1763 spin_lock(&lli->lli_lock);
1764 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1765 spin_unlock(&lli->lli_lock);
1766 CWARN("no group lock held\n");
1769 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1771 if (fd->fd_grouplock.cg_gid != arg) {
1772 CWARN("group lock %lu doesn't match current id %lu\n",
1773 arg, fd->fd_grouplock.cg_gid);
1774 spin_unlock(&lli->lli_lock);
1778 grouplock = fd->fd_grouplock;
1779 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1780 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1781 spin_unlock(&lli->lli_lock);
1783 cl_put_grouplock(&grouplock);
1784 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1789 * Close inode open handle
1791 * \param dentry [in] dentry which contains the inode
1792 * \param it [in,out] intent which contains open info and result
1795 * \retval <0 failure
1797 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
1799 struct inode *inode = dentry->d_inode;
1800 struct obd_client_handle *och;
1806 /* Root ? Do nothing. */
1807 if (dentry->d_inode->i_sb->s_root == dentry)
1810 /* No open handle to close? Move away */
1811 if (!it_disposition(it, DISP_OPEN_OPEN))
1814 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1816 OBD_ALLOC(och, sizeof(*och));
1818 GOTO(out, rc = -ENOMEM);
1820 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
1822 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1825 /* this one is in place of ll_file_open */
1826 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
1827 ptlrpc_req_finished(it->d.lustre.it_data);
1828 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1834 * Get size for inode for which FIEMAP mapping is requested.
1835 * Make the FIEMAP get_info call and returns the result.
1837 static int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1840 struct obd_export *exp = ll_i2dtexp(inode);
1841 struct lov_stripe_md *lsm = NULL;
1842 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1843 __u32 vallen = num_bytes;
1847 /* Checks for fiemap flags */
1848 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1849 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1853 /* Check for FIEMAP_FLAG_SYNC */
1854 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1855 rc = filemap_fdatawrite(inode->i_mapping);
1860 lsm = ccc_inode_lsm_get(inode);
1864 /* If the stripe_count > 1 and the application does not understand
1865 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1867 if (lsm->lsm_stripe_count > 1 &&
1868 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER))
1869 GOTO(out, rc = -EOPNOTSUPP);
1871 fm_key.oa.o_oi = lsm->lsm_oi;
1872 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1874 obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE);
1875 obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid);
1876 /* If filesize is 0, then there would be no objects for mapping */
1877 if (fm_key.oa.o_size == 0) {
1878 fiemap->fm_mapped_extents = 0;
1882 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1884 rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen,
1887 CERROR("obd_get_info failed: rc = %d\n", rc);
1890 ccc_inode_lsm_put(inode, lsm);
1894 int ll_fid2path(struct inode *inode, void *arg)
1896 struct obd_export *exp = ll_i2mdexp(inode);
1897 struct getinfo_fid2path *gfout, *gfin;
1901 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1902 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
1905 /* Need to get the buflen */
1906 OBD_ALLOC_PTR(gfin);
1909 if (copy_from_user(gfin, arg, sizeof(*gfin))) {
1914 outsize = sizeof(*gfout) + gfin->gf_pathlen;
1915 OBD_ALLOC(gfout, outsize);
1916 if (gfout == NULL) {
1920 memcpy(gfout, gfin, sizeof(*gfout));
1923 /* Call mdc_iocontrol */
1924 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1928 if (copy_to_user(arg, gfout, outsize))
1932 OBD_FREE(gfout, outsize);
1936 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1938 struct ll_user_fiemap *fiemap_s;
1939 size_t num_bytes, ret_bytes;
1940 unsigned int extent_count;
1943 /* Get the extent count so we can calculate the size of
1944 * required fiemap buffer */
1945 if (get_user(extent_count,
1946 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1948 num_bytes = sizeof(*fiemap_s) + (extent_count *
1949 sizeof(struct ll_fiemap_extent));
1951 OBD_ALLOC_LARGE(fiemap_s, num_bytes);
1952 if (fiemap_s == NULL)
1955 /* get the fiemap value */
1956 if (copy_from_user(fiemap_s, (struct ll_user_fiemap __user *)arg,
1958 GOTO(error, rc = -EFAULT);
1960 /* If fm_extent_count is non-zero, read the first extent since
1961 * it is used to calculate end_offset and device from previous
1964 if (copy_from_user(&fiemap_s->fm_extents[0],
1965 (char __user *)arg + sizeof(*fiemap_s),
1966 sizeof(struct ll_fiemap_extent)))
1967 GOTO(error, rc = -EFAULT);
1970 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1974 ret_bytes = sizeof(struct ll_user_fiemap);
1976 if (extent_count != 0)
1977 ret_bytes += (fiemap_s->fm_mapped_extents *
1978 sizeof(struct ll_fiemap_extent));
1980 if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
1984 OBD_FREE_LARGE(fiemap_s, num_bytes);
1989 * Read the data_version for inode.
1991 * This value is computed using stripe object version on OST.
1992 * Version is computed using server side locking.
1994 * @param sync if do sync on the OST side;
1996 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
1997 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
1999 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2001 struct lov_stripe_md *lsm = NULL;
2002 struct ll_sb_info *sbi = ll_i2sbi(inode);
2003 struct obdo *obdo = NULL;
2007 /* If no stripe, we consider version is 0. */
2008 lsm = ccc_inode_lsm_get(inode);
2009 if (!lsm_has_objects(lsm)) {
2011 CDEBUG(D_INODE, "No object for inode\n");
2015 OBD_ALLOC_PTR(obdo);
2017 GOTO(out, rc = -ENOMEM);
2019 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, flags);
2021 if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
2024 *data_version = obdo->o_data_version;
2030 ccc_inode_lsm_put(inode, lsm);
2035 * Trigger a HSM release request for the provided inode.
2037 int ll_hsm_release(struct inode *inode)
2039 struct cl_env_nest nest;
2041 struct obd_client_handle *och = NULL;
2042 __u64 data_version = 0;
2046 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2047 ll_get_fsname(inode->i_sb, NULL, 0),
2048 PFID(&ll_i2info(inode)->lli_fid));
2050 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2052 GOTO(out, rc = PTR_ERR(och));
2054 /* Grab latest data_version and [am]time values */
2055 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2059 env = cl_env_nested_get(&nest);
2061 GOTO(out, rc = PTR_ERR(env));
2063 ll_merge_lvb(env, inode);
2064 cl_env_nested_put(&nest, env);
2066 /* Release the file.
2067 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2068 * we still need it to pack l_remote_handle to MDT. */
2069 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp, inode, och,
2075 if (och != NULL && !IS_ERR(och)) /* close the file */
2076 ll_lease_close(och, inode, NULL);
2081 struct ll_swap_stack {
2082 struct iattr ia1, ia2;
2084 struct inode *inode1, *inode2;
2085 bool check_dv1, check_dv2;
2088 static int ll_swap_layouts(struct file *file1, struct file *file2,
2089 struct lustre_swap_layouts *lsl)
2091 struct mdc_swap_layouts msl;
2092 struct md_op_data *op_data;
2095 struct ll_swap_stack *llss = NULL;
2098 OBD_ALLOC_PTR(llss);
2102 llss->inode1 = file1->f_dentry->d_inode;
2103 llss->inode2 = file2->f_dentry->d_inode;
2105 if (!S_ISREG(llss->inode2->i_mode))
2106 GOTO(free, rc = -EINVAL);
2108 if (inode_permission(llss->inode1, MAY_WRITE) ||
2109 inode_permission(llss->inode2, MAY_WRITE))
2110 GOTO(free, rc = -EPERM);
2112 if (llss->inode2->i_sb != llss->inode1->i_sb)
2113 GOTO(free, rc = -EXDEV);
2115 /* we use 2 bool because it is easier to swap than 2 bits */
2116 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2117 llss->check_dv1 = true;
2119 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2120 llss->check_dv2 = true;
2122 /* we cannot use lsl->sl_dvX directly because we may swap them */
2123 llss->dv1 = lsl->sl_dv1;
2124 llss->dv2 = lsl->sl_dv2;
2126 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2127 if (rc == 0) /* same file, done! */
2130 if (rc < 0) { /* sequentialize it */
2131 swap(llss->inode1, llss->inode2);
2133 swap(llss->dv1, llss->dv2);
2134 swap(llss->check_dv1, llss->check_dv2);
2138 if (gid != 0) { /* application asks to flush dirty cache */
2139 rc = ll_get_grouplock(llss->inode1, file1, gid);
2143 rc = ll_get_grouplock(llss->inode2, file2, gid);
2145 ll_put_grouplock(llss->inode1, file1, gid);
2150 /* to be able to restore mtime and atime after swap
2151 * we need to first save them */
2153 (SWAP_LAYOUTS_KEEP_MTIME | SWAP_LAYOUTS_KEEP_ATIME)) {
2154 llss->ia1.ia_mtime = llss->inode1->i_mtime;
2155 llss->ia1.ia_atime = llss->inode1->i_atime;
2156 llss->ia1.ia_valid = ATTR_MTIME | ATTR_ATIME;
2157 llss->ia2.ia_mtime = llss->inode2->i_mtime;
2158 llss->ia2.ia_atime = llss->inode2->i_atime;
2159 llss->ia2.ia_valid = ATTR_MTIME | ATTR_ATIME;
2162 /* ultimate check, before swaping the layouts we check if
2163 * dataversion has changed (if requested) */
2164 if (llss->check_dv1) {
2165 rc = ll_data_version(llss->inode1, &dv, 0);
2168 if (dv != llss->dv1)
2169 GOTO(putgl, rc = -EAGAIN);
2172 if (llss->check_dv2) {
2173 rc = ll_data_version(llss->inode2, &dv, 0);
2176 if (dv != llss->dv2)
2177 GOTO(putgl, rc = -EAGAIN);
2180 /* struct md_op_data is used to send the swap args to the mdt
2181 * only flags is missing, so we use struct mdc_swap_layouts
2182 * through the md_op_data->op_data */
2183 /* flags from user space have to be converted before they are send to
2184 * server, no flag is sent today, they are only used on the client */
2187 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2188 0, LUSTRE_OPC_ANY, &msl);
2189 if (IS_ERR(op_data))
2190 GOTO(free, rc = PTR_ERR(op_data));
2192 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2193 sizeof(*op_data), op_data, NULL);
2194 ll_finish_md_op_data(op_data);
2198 ll_put_grouplock(llss->inode2, file2, gid);
2199 ll_put_grouplock(llss->inode1, file1, gid);
2202 /* rc can be set from obd_iocontrol() or from a GOTO(putgl, ...) */
2206 /* clear useless flags */
2207 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_MTIME)) {
2208 llss->ia1.ia_valid &= ~ATTR_MTIME;
2209 llss->ia2.ia_valid &= ~ATTR_MTIME;
2212 if (!(lsl->sl_flags & SWAP_LAYOUTS_KEEP_ATIME)) {
2213 llss->ia1.ia_valid &= ~ATTR_ATIME;
2214 llss->ia2.ia_valid &= ~ATTR_ATIME;
2217 /* update time if requested */
2219 if (llss->ia2.ia_valid != 0) {
2220 mutex_lock(&llss->inode1->i_mutex);
2221 rc = ll_setattr(file1->f_dentry, &llss->ia2);
2222 mutex_unlock(&llss->inode1->i_mutex);
2225 if (llss->ia1.ia_valid != 0) {
2228 mutex_lock(&llss->inode2->i_mutex);
2229 rc1 = ll_setattr(file2->f_dentry, &llss->ia1);
2230 mutex_unlock(&llss->inode2->i_mutex);
2242 static int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2244 struct md_op_data *op_data;
2247 /* Non-root users are forbidden to set or clear flags which are
2248 * NOT defined in HSM_USER_MASK. */
2249 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2250 !cfs_capable(CFS_CAP_SYS_ADMIN))
2253 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2254 LUSTRE_OPC_ANY, hss);
2255 if (IS_ERR(op_data))
2256 RETURN(PTR_ERR(op_data));
2258 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, ll_i2mdexp(inode),
2259 sizeof(*op_data), op_data, NULL);
2261 ll_finish_md_op_data(op_data);
2266 static int ll_hsm_import(struct inode *inode, struct file *file,
2267 struct hsm_user_import *hui)
2269 struct hsm_state_set *hss = NULL;
2270 struct iattr *attr = NULL;
2274 if (!S_ISREG(inode->i_mode))
2280 GOTO(out, rc = -ENOMEM);
2282 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2283 hss->hss_archive_id = hui->hui_archive_id;
2284 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2285 rc = ll_hsm_state_set(inode, hss);
2289 OBD_ALLOC_PTR(attr);
2291 GOTO(out, rc = -ENOMEM);
2293 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2294 attr->ia_mode |= S_IFREG;
2295 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2296 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2297 attr->ia_size = hui->hui_size;
2298 attr->ia_mtime.tv_sec = hui->hui_mtime;
2299 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2300 attr->ia_atime.tv_sec = hui->hui_atime;
2301 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2303 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2304 ATTR_UID | ATTR_GID |
2305 ATTR_MTIME | ATTR_MTIME_SET |
2306 ATTR_ATIME | ATTR_ATIME_SET;
2308 rc = ll_setattr_raw(file->f_dentry, attr, true);
2323 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2325 struct inode *inode = file->f_dentry->d_inode;
2326 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2330 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
2331 PFID(ll_inode2fid(inode)), inode, cmd);
2332 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2334 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2335 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2339 case LL_IOC_GETFLAGS:
2340 /* Get the current value of the file flags */
2341 return put_user(fd->fd_flags, (int *)arg);
2342 case LL_IOC_SETFLAGS:
2343 case LL_IOC_CLRFLAGS:
2344 /* Set or clear specific file flags */
2345 /* XXX This probably needs checks to ensure the flags are
2346 * not abused, and to handle any flag side effects.
2348 if (get_user(flags, (int *) arg))
2351 if (cmd == LL_IOC_SETFLAGS) {
2352 if ((flags & LL_FILE_IGNORE_LOCK) &&
2353 !(file->f_flags & O_DIRECT)) {
2354 CERROR("%s: unable to disable locking on "
2355 "non-O_DIRECT file\n", current->comm);
2359 fd->fd_flags |= flags;
2361 fd->fd_flags &= ~flags;
2364 case LL_IOC_LOV_SETSTRIPE:
2365 RETURN(ll_lov_setstripe(inode, file, arg));
2366 case LL_IOC_LOV_SETEA:
2367 RETURN(ll_lov_setea(inode, file, arg));
2368 case LL_IOC_LOV_SWAP_LAYOUTS: {
2370 struct lustre_swap_layouts lsl;
2372 if (copy_from_user(&lsl, (char *)arg,
2373 sizeof(struct lustre_swap_layouts)))
2376 if ((file->f_flags & O_ACCMODE) == 0) /* O_RDONLY */
2379 file2 = fget(lsl.sl_fd);
2384 if ((file2->f_flags & O_ACCMODE) != 0) /* O_WRONLY or O_RDWR */
2385 rc = ll_swap_layouts(file, file2, &lsl);
2389 case LL_IOC_LOV_GETSTRIPE:
2390 RETURN(ll_lov_getstripe(inode, arg));
2391 case LL_IOC_RECREATE_OBJ:
2392 RETURN(ll_lov_recreate_obj(inode, arg));
2393 case LL_IOC_RECREATE_FID:
2394 RETURN(ll_lov_recreate_fid(inode, arg));
2395 case FSFILT_IOC_FIEMAP:
2396 RETURN(ll_ioctl_fiemap(inode, arg));
2397 case FSFILT_IOC_GETFLAGS:
2398 case FSFILT_IOC_SETFLAGS:
2399 RETURN(ll_iocontrol(inode, file, cmd, arg));
2400 case FSFILT_IOC_GETVERSION_OLD:
2401 case FSFILT_IOC_GETVERSION:
2402 RETURN(put_user(inode->i_generation, (int *)arg));
2403 case LL_IOC_GROUP_LOCK:
2404 RETURN(ll_get_grouplock(inode, file, arg));
2405 case LL_IOC_GROUP_UNLOCK:
2406 RETURN(ll_put_grouplock(inode, file, arg));
2407 case IOC_OBD_STATFS:
2408 RETURN(ll_obd_statfs(inode, (void *)arg));
2410 /* We need to special case any other ioctls we want to handle,
2411 * to send them to the MDS/OST as appropriate and to properly
2412 * network encode the arg field.
2413 case FSFILT_IOC_SETVERSION_OLD:
2414 case FSFILT_IOC_SETVERSION:
2416 case LL_IOC_FLUSHCTX:
2417 RETURN(ll_flush_ctx(inode));
2418 case LL_IOC_PATH2FID: {
2419 if (copy_to_user((void *)arg, ll_inode2fid(inode),
2420 sizeof(struct lu_fid)))
2425 case OBD_IOC_FID2PATH:
2426 RETURN(ll_fid2path(inode, (void *)arg));
2427 case LL_IOC_DATA_VERSION: {
2428 struct ioc_data_version idv;
2431 if (copy_from_user(&idv, (char *)arg, sizeof(idv)))
2434 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
2435 rc = ll_data_version(inode, &idv.idv_version, idv.idv_flags);
2437 if (rc == 0 && copy_to_user((char *) arg, &idv, sizeof(idv)))
2443 case LL_IOC_GET_MDTIDX: {
2446 mdtidx = ll_get_mdt_idx(inode);
2450 if (put_user((int)mdtidx, (int*)arg))
2455 case OBD_IOC_GETDTNAME:
2456 case OBD_IOC_GETMDNAME:
2457 RETURN(ll_get_obd_name(inode, cmd, arg));
2458 case LL_IOC_HSM_STATE_GET: {
2459 struct md_op_data *op_data;
2460 struct hsm_user_state *hus;
2467 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2468 LUSTRE_OPC_ANY, hus);
2469 if (IS_ERR(op_data)) {
2471 RETURN(PTR_ERR(op_data));
2474 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2477 if (copy_to_user((void *)arg, hus, sizeof(*hus)))
2480 ll_finish_md_op_data(op_data);
2484 case LL_IOC_HSM_STATE_SET: {
2485 struct hsm_state_set *hss;
2492 if (copy_from_user(hss, (char *)arg, sizeof(*hss))) {
2497 rc = ll_hsm_state_set(inode, hss);
2502 case LL_IOC_HSM_ACTION: {
2503 struct md_op_data *op_data;
2504 struct hsm_current_action *hca;
2511 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2512 LUSTRE_OPC_ANY, hca);
2513 if (IS_ERR(op_data)) {
2515 RETURN(PTR_ERR(op_data));
2518 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
2521 if (copy_to_user((char *)arg, hca, sizeof(*hca)))
2524 ll_finish_md_op_data(op_data);
2528 case LL_IOC_SET_LEASE: {
2529 struct ll_inode_info *lli = ll_i2info(inode);
2530 struct obd_client_handle *och = NULL;
2536 if (!(file->f_mode & FMODE_WRITE))
2541 if (!(file->f_mode & FMODE_READ))
2546 mutex_lock(&lli->lli_och_mutex);
2547 if (fd->fd_lease_och != NULL) {
2548 och = fd->fd_lease_och;
2549 fd->fd_lease_och = NULL;
2551 mutex_unlock(&lli->lli_och_mutex);
2554 mode = och->och_flags &(FMODE_READ|FMODE_WRITE);
2555 rc = ll_lease_close(och, inode, &lease_broken);
2556 if (rc == 0 && lease_broken)
2562 /* return the type of lease or error */
2563 RETURN(rc < 0 ? rc : (int)mode);
2568 CDEBUG(D_INODE, "Set lease with mode %d\n", mode);
2570 /* apply for lease */
2571 och = ll_lease_open(inode, file, mode, 0);
2573 RETURN(PTR_ERR(och));
2576 mutex_lock(&lli->lli_och_mutex);
2577 if (fd->fd_lease_och == NULL) {
2578 fd->fd_lease_och = och;
2581 mutex_unlock(&lli->lli_och_mutex);
2583 /* impossible now that only excl is supported for now */
2584 ll_lease_close(och, inode, &lease_broken);
2589 case LL_IOC_GET_LEASE: {
2590 struct ll_inode_info *lli = ll_i2info(inode);
2591 struct ldlm_lock *lock = NULL;
2594 mutex_lock(&lli->lli_och_mutex);
2595 if (fd->fd_lease_och != NULL) {
2596 struct obd_client_handle *och = fd->fd_lease_och;
2598 lock = ldlm_handle2lock(&och->och_lease_handle);
2600 lock_res_and_lock(lock);
2601 if (!ldlm_is_cancel(lock))
2602 rc = och->och_flags &
2603 (FMODE_READ | FMODE_WRITE);
2604 unlock_res_and_lock(lock);
2605 LDLM_LOCK_PUT(lock);
2608 mutex_unlock(&lli->lli_och_mutex);
2611 case LL_IOC_HSM_IMPORT: {
2612 struct hsm_user_import *hui;
2618 if (copy_from_user(hui, (void *)arg, sizeof(*hui))) {
2623 rc = ll_hsm_import(inode, file, hui);
2633 ll_iocontrol_call(inode, file, cmd, arg, &err))
2636 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2642 #ifndef HAVE_FILE_LLSEEK_SIZE
2643 static inline loff_t
2644 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
2646 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2648 if (offset > maxsize)
2651 if (offset != file->f_pos) {
2652 file->f_pos = offset;
2653 file->f_version = 0;
2659 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
2660 loff_t maxsize, loff_t eof)
2662 struct inode *inode = file->f_dentry->d_inode;
2670 * Here we special-case the lseek(fd, 0, SEEK_CUR)
2671 * position-querying operation. Avoid rewriting the "same"
2672 * f_pos value back to the file because a concurrent read(),
2673 * write() or lseek() might have altered it
2678 * f_lock protects against read/modify/write race with other
2679 * SEEK_CURs. Note that parallel writes and reads behave
2682 mutex_lock(&inode->i_mutex);
2683 offset = llseek_execute(file, file->f_pos + offset, maxsize);
2684 mutex_unlock(&inode->i_mutex);
2688 * In the generic case the entire file is data, so as long as
2689 * offset isn't at the end of the file then the offset is data.
2696 * There is a virtual hole at the end of the file, so as long as
2697 * offset isn't i_size or larger, return i_size.
2705 return llseek_execute(file, offset, maxsize);
2709 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2711 struct inode *inode = file->f_dentry->d_inode;
2712 loff_t retval, eof = 0;
2715 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
2716 (origin == SEEK_CUR) ? file->f_pos : 0);
2717 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
2718 PFID(ll_inode2fid(inode)), inode, retval, retval,
2720 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2722 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
2723 retval = ll_glimpse_size(inode);
2726 eof = i_size_read(inode);
2729 retval = ll_generic_file_llseek_size(file, offset, origin,
2730 ll_file_maxbytes(inode), eof);
2734 static int ll_flush(struct file *file, fl_owner_t id)
2736 struct inode *inode = file->f_dentry->d_inode;
2737 struct ll_inode_info *lli = ll_i2info(inode);
2738 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2741 LASSERT(!S_ISDIR(inode->i_mode));
2743 /* catch async errors that were recorded back when async writeback
2744 * failed for pages in this mapping. */
2745 rc = lli->lli_async_rc;
2746 lli->lli_async_rc = 0;
2747 if (lli->lli_clob != NULL) {
2748 err = lov_read_and_clear_async_rc(lli->lli_clob);
2753 /* The application has been told write failure already.
2754 * Do not report failure again. */
2755 if (fd->fd_write_failed)
2757 return rc ? -EIO : 0;
2761 * Called to make sure a portion of file has been written out.
2762 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
2764 * Return how many pages have been written.
2766 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
2767 enum cl_fsync_mode mode, int ignore_layout)
2769 struct cl_env_nest nest;
2772 struct obd_capa *capa = NULL;
2773 struct cl_fsync_io *fio;
2777 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
2778 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
2781 env = cl_env_nested_get(&nest);
2783 RETURN(PTR_ERR(env));
2785 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
2787 io = ccc_env_thread_io(env);
2788 io->ci_obj = cl_i2info(inode)->lli_clob;
2789 io->ci_ignore_layout = ignore_layout;
2791 /* initialize parameters for sync */
2792 fio = &io->u.ci_fsync;
2793 fio->fi_capa = capa;
2794 fio->fi_start = start;
2796 fio->fi_fid = ll_inode2fid(inode);
2797 fio->fi_mode = mode;
2798 fio->fi_nr_written = 0;
2800 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
2801 result = cl_io_loop(env, io);
2803 result = io->ci_result;
2805 result = fio->fi_nr_written;
2806 cl_io_fini(env, io);
2807 cl_env_nested_put(&nest, env);
2815 * When dentry is provided (the 'else' case), *file->f_dentry may be
2816 * null and dentry must be used directly rather than pulled from
2817 * *file->f_dentry as is done otherwise.
2820 #ifdef HAVE_FILE_FSYNC_4ARGS
2821 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2823 struct dentry *dentry = file->f_dentry;
2824 #elif defined(HAVE_FILE_FSYNC_2ARGS)
2825 int ll_fsync(struct file *file, int datasync)
2827 struct dentry *dentry = file->f_dentry;
2829 loff_t end = LLONG_MAX;
2831 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
2834 loff_t end = LLONG_MAX;
2836 struct inode *inode = dentry->d_inode;
2837 struct ll_inode_info *lli = ll_i2info(inode);
2838 struct ptlrpc_request *req;
2839 struct obd_capa *oc;
2843 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2844 PFID(ll_inode2fid(inode)), inode);
2845 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2847 #ifdef HAVE_FILE_FSYNC_4ARGS
2848 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2849 mutex_lock(&inode->i_mutex);
2851 /* fsync's caller has already called _fdata{sync,write}, we want
2852 * that IO to finish before calling the osc and mdc sync methods */
2853 rc = filemap_fdatawait(inode->i_mapping);
2856 /* catch async errors that were recorded back when async writeback
2857 * failed for pages in this mapping. */
2858 if (!S_ISDIR(inode->i_mode)) {
2859 err = lli->lli_async_rc;
2860 lli->lli_async_rc = 0;
2863 err = lov_read_and_clear_async_rc(lli->lli_clob);
2868 oc = ll_mdscapa_get(inode);
2869 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2875 ptlrpc_req_finished(req);
2877 if (S_ISREG(inode->i_mode)) {
2878 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2880 err = cl_sync_file_range(inode, start, end, CL_FSYNC_ALL, 0);
2881 if (rc == 0 && err < 0)
2884 fd->fd_write_failed = true;
2886 fd->fd_write_failed = false;
2889 #ifdef HAVE_FILE_FSYNC_4ARGS
2890 mutex_unlock(&inode->i_mutex);
2896 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2898 struct inode *inode = file->f_dentry->d_inode;
2899 struct ll_sb_info *sbi = ll_i2sbi(inode);
2900 struct ldlm_enqueue_info einfo = {
2901 .ei_type = LDLM_FLOCK,
2902 .ei_cb_cp = ldlm_flock_completion_ast,
2903 .ei_cbdata = file_lock,
2905 struct md_op_data *op_data;
2906 struct lustre_handle lockh = {0};
2907 ldlm_policy_data_t flock = {{0}};
2908 int fl_type = file_lock->fl_type;
2914 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
2915 PFID(ll_inode2fid(inode)), file_lock);
2917 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2919 if (file_lock->fl_flags & FL_FLOCK) {
2920 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2921 /* flocks are whole-file locks */
2922 flock.l_flock.end = OFFSET_MAX;
2923 /* For flocks owner is determined by the local file desctiptor*/
2924 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
2925 } else if (file_lock->fl_flags & FL_POSIX) {
2926 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
2927 flock.l_flock.start = file_lock->fl_start;
2928 flock.l_flock.end = file_lock->fl_end;
2932 flock.l_flock.pid = file_lock->fl_pid;
2934 /* Somewhat ugly workaround for svc lockd.
2935 * lockd installs custom fl_lmops->lm_compare_owner that checks
2936 * for the fl_owner to be the same (which it always is on local node
2937 * I guess between lockd processes) and then compares pid.
2938 * As such we assign pid to the owner field to make it all work,
2939 * conflict with normal locks is unlikely since pid space and
2940 * pointer space for current->files are not intersecting */
2941 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
2942 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
2946 einfo.ei_mode = LCK_PR;
2949 /* An unlock request may or may not have any relation to
2950 * existing locks so we may not be able to pass a lock handle
2951 * via a normal ldlm_lock_cancel() request. The request may even
2952 * unlock a byte range in the middle of an existing lock. In
2953 * order to process an unlock request we need all of the same
2954 * information that is given with a normal read or write record
2955 * lock request. To avoid creating another ldlm unlock (cancel)
2956 * message we'll treat a LCK_NL flock request as an unlock. */
2957 einfo.ei_mode = LCK_NL;
2960 einfo.ei_mode = LCK_PW;
2963 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
2978 flags = LDLM_FL_BLOCK_NOWAIT;
2984 flags = LDLM_FL_TEST_LOCK;
2987 CERROR("unknown fcntl lock command: %d\n", cmd);
2991 /* Save the old mode so that if the mode in the lock changes we
2992 * can decrement the appropriate reader or writer refcount. */
2993 file_lock->fl_type = einfo.ei_mode;
2995 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2996 LUSTRE_OPC_ANY, NULL);
2997 if (IS_ERR(op_data))
2998 RETURN(PTR_ERR(op_data));
3000 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags="LPX64", mode=%u, "
3001 "start="LPU64", end="LPU64"\n", PFID(ll_inode2fid(inode)),
3002 flock.l_flock.pid, flags, einfo.ei_mode,
3003 flock.l_flock.start, flock.l_flock.end);
3005 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data, &lockh,
3008 /* Restore the file lock type if not TEST lock. */
3009 if (!(flags & LDLM_FL_TEST_LOCK))
3010 file_lock->fl_type = fl_type;
3012 if ((file_lock->fl_flags & FL_FLOCK) &&
3013 (rc == 0 || file_lock->fl_type == F_UNLCK))
3014 rc2 = flock_lock_file_wait(file, file_lock);
3015 if ((file_lock->fl_flags & FL_POSIX) &&
3016 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
3017 !(flags & LDLM_FL_TEST_LOCK))
3018 rc2 = posix_lock_file_wait(file, file_lock);
3020 if (rc2 && file_lock->fl_type != F_UNLCK) {
3021 einfo.ei_mode = LCK_NL;
3022 md_enqueue(sbi->ll_md_exp, &einfo, &flock, NULL, op_data,
3027 ll_finish_md_op_data(op_data);
3032 int ll_get_fid_by_name(struct inode *parent, const char *name,
3033 int namelen, struct lu_fid *fid)
3035 struct md_op_data *op_data = NULL;
3036 struct mdt_body *body;
3037 struct ptlrpc_request *req;
3041 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
3042 LUSTRE_OPC_ANY, NULL);
3043 if (IS_ERR(op_data))
3044 RETURN(PTR_ERR(op_data));
3046 op_data->op_valid = OBD_MD_FLID;
3047 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
3048 ll_finish_md_op_data(op_data);
3052 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3054 GOTO(out_req, rc = -EFAULT);
3056 *fid = body->mbo_fid1;
3058 ptlrpc_req_finished(req);
3062 int ll_migrate(struct inode *parent, struct file *file, int mdtidx,
3063 const char *name, int namelen)
3065 struct dentry *dchild = NULL;
3066 struct inode *child_inode = NULL;
3067 struct md_op_data *op_data;
3068 struct ptlrpc_request *request = NULL;
3073 CDEBUG(D_VFSTRACE, "migrate %s under "DFID" to MDT%04x\n",
3074 name, PFID(ll_inode2fid(parent)), mdtidx);
3076 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
3077 0, LUSTRE_OPC_ANY, NULL);
3078 if (IS_ERR(op_data))
3079 RETURN(PTR_ERR(op_data));
3081 /* Get child FID first */
3082 qstr.hash = full_name_hash(name, namelen);
3085 dchild = d_lookup(file->f_dentry, &qstr);
3086 if (dchild != NULL && dchild->d_inode != NULL) {
3087 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
3088 if (dchild->d_inode != NULL) {
3089 child_inode = igrab(dchild->d_inode);
3090 ll_invalidate_aliases(child_inode);
3094 rc = ll_get_fid_by_name(parent, name, namelen,
3100 if (!fid_is_sane(&op_data->op_fid3)) {
3101 CERROR("%s: migrate %s , but fid "DFID" is insane\n",
3102 ll_get_fsname(parent->i_sb, NULL, 0), name,
3103 PFID(&op_data->op_fid3));
3104 GOTO(out_free, rc = -EINVAL);
3107 rc = ll_get_mdt_idx_by_fid(ll_i2sbi(parent), &op_data->op_fid3);
3112 CDEBUG(D_INFO, "%s:"DFID" is already on MDT%d.\n", name,
3113 PFID(&op_data->op_fid3), mdtidx);
3114 GOTO(out_free, rc = 0);
3117 op_data->op_mds = mdtidx;
3118 op_data->op_cli_flags = CLI_MIGRATE;
3119 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name,
3120 namelen, name, namelen, &request);
3122 ll_update_times(request, parent);
3124 ptlrpc_req_finished(request);
3129 if (child_inode != NULL) {
3130 clear_nlink(child_inode);
3134 ll_finish_md_op_data(op_data);
3139 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
3147 * test if some locks matching bits and l_req_mode are acquired
3148 * - bits can be in different locks
3149 * - if found clear the common lock bits in *bits
3150 * - the bits not found, are kept in *bits
3152 * \param bits [IN] searched lock bits [IN]
3153 * \param l_req_mode [IN] searched lock mode
3154 * \retval boolean, true iff all bits are found
3156 int ll_have_md_lock(struct inode *inode, __u64 *bits, ldlm_mode_t l_req_mode)
3158 struct lustre_handle lockh;
3159 ldlm_policy_data_t policy;
3160 ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ?
3161 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
3170 fid = &ll_i2info(inode)->lli_fid;
3171 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
3172 ldlm_lockname[mode]);
3174 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
3175 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
3176 policy.l_inodebits.bits = *bits & (1 << i);
3177 if (policy.l_inodebits.bits == 0)
3180 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
3181 &policy, mode, &lockh)) {
3182 struct ldlm_lock *lock;
3184 lock = ldlm_handle2lock(&lockh);
3187 ~(lock->l_policy_data.l_inodebits.bits);
3188 LDLM_LOCK_PUT(lock);
3190 *bits &= ~policy.l_inodebits.bits;
3197 ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
3198 struct lustre_handle *lockh, __u64 flags,
3201 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
3206 fid = &ll_i2info(inode)->lli_fid;
3207 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
3209 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
3210 fid, LDLM_IBITS, &policy, mode, lockh);
3215 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
3217 /* Already unlinked. Just update nlink and return success */
3218 if (rc == -ENOENT) {
3220 /* This path cannot be hit for regular files unless in
3221 * case of obscure races, so no need to to validate
3223 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
3225 } else if (rc != 0) {
3226 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
3227 "%s: revalidate FID "DFID" error: rc = %d\n",
3228 ll_get_fsname(inode->i_sb, NULL, 0),
3229 PFID(ll_inode2fid(inode)), rc);
3235 static int __ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3237 struct inode *inode = dentry->d_inode;
3238 struct ptlrpc_request *req = NULL;
3239 struct obd_export *exp;
3243 LASSERT(inode != NULL);
3245 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
3246 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
3248 exp = ll_i2mdexp(inode);
3250 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
3251 * But under CMD case, it caused some lock issues, should be fixed
3252 * with new CMD ibits lock. See bug 12718 */
3253 if (exp_connect_flags(exp) & OBD_CONNECT_ATTRFID) {
3254 struct lookup_intent oit = { .it_op = IT_GETATTR };
3255 struct md_op_data *op_data;
3257 if (ibits == MDS_INODELOCK_LOOKUP)
3258 oit.it_op = IT_LOOKUP;
3260 /* Call getattr by fid, so do not provide name at all. */
3261 op_data = ll_prep_md_op_data(NULL, dentry->d_inode,
3262 dentry->d_inode, NULL, 0, 0,
3263 LUSTRE_OPC_ANY, NULL);
3264 if (IS_ERR(op_data))
3265 RETURN(PTR_ERR(op_data));
3267 oit.it_create_mode |= M_CHECK_STALE;
3268 rc = md_intent_lock(exp, op_data, &oit, &req,
3269 &ll_md_blocking_ast, 0);
3270 ll_finish_md_op_data(op_data);
3271 oit.it_create_mode &= ~M_CHECK_STALE;
3273 rc = ll_inode_revalidate_fini(inode, rc);
3277 rc = ll_revalidate_it_finish(req, &oit, dentry);
3279 ll_intent_release(&oit);
3283 /* Unlinked? Unhash dentry, so it is not picked up later by
3284 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
3285 here to preserve get_cwd functionality on 2.6.
3287 if (!dentry->d_inode->i_nlink)
3288 d_lustre_invalidate(dentry, 0);
3290 ll_lookup_finish_locks(&oit, dentry);
3291 } else if (!ll_have_md_lock(dentry->d_inode, &ibits, LCK_MINMODE)) {
3292 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
3293 obd_valid valid = OBD_MD_FLGETATTR;
3294 struct md_op_data *op_data;
3297 if (S_ISREG(inode->i_mode)) {
3298 rc = ll_get_default_mdsize(sbi, &ealen);
3301 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
3304 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
3305 0, ealen, LUSTRE_OPC_ANY,
3307 if (IS_ERR(op_data))
3308 RETURN(PTR_ERR(op_data));
3310 op_data->op_valid = valid;
3311 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
3312 * capa for this inode. Because we only keep capas of dirs
3314 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
3315 ll_finish_md_op_data(op_data);
3317 rc = ll_inode_revalidate_fini(inode, rc);
3321 rc = ll_prep_inode(&inode, req, NULL, NULL);
3324 ptlrpc_req_finished(req);
3328 static int ll_merge_md_attr(struct inode *inode)
3330 struct cl_attr attr = { 0 };
3333 LASSERT(ll_i2info(inode)->lli_lsm_md != NULL);
3334 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
3339 ll_i2info(inode)->lli_stripe_dir_size = attr.cat_size;
3340 ll_i2info(inode)->lli_stripe_dir_nlink = attr.cat_nlink;
3342 ll_i2info(inode)->lli_lvb.lvb_atime = attr.cat_atime;
3343 ll_i2info(inode)->lli_lvb.lvb_mtime = attr.cat_mtime;
3344 ll_i2info(inode)->lli_lvb.lvb_ctime = attr.cat_ctime;
3350 ll_inode_revalidate(struct dentry *dentry, __u64 ibits)
3352 struct inode *inode = dentry->d_inode;
3356 rc = __ll_inode_revalidate(dentry, ibits);
3360 /* if object isn't regular file, don't validate size */
3361 if (!S_ISREG(inode->i_mode)) {
3362 if (S_ISDIR(inode->i_mode) &&
3363 ll_i2info(inode)->lli_lsm_md != NULL) {
3364 rc = ll_merge_md_attr(inode);
3369 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime;
3370 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime;
3371 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime;
3373 /* In case of restore, the MDT has the right size and has
3374 * already send it back without granting the layout lock,
3375 * inode is up-to-date so glimpse is useless.
3376 * Also to glimpse we need the layout, in case of a running
3377 * restore the MDT holds the layout lock so the glimpse will
3378 * block up to the end of restore (getattr will block)
3380 if (!(ll_i2info(inode)->lli_flags & LLIF_FILE_RESTORING))
3381 rc = ll_glimpse_size(inode);
3386 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
3388 struct inode *inode = de->d_inode;
3389 struct ll_sb_info *sbi = ll_i2sbi(inode);
3390 struct ll_inode_info *lli = ll_i2info(inode);
3393 res = ll_inode_revalidate(de, MDS_INODELOCK_UPDATE |
3394 MDS_INODELOCK_LOOKUP);
3395 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
3400 stat->dev = inode->i_sb->s_dev;
3401 if (ll_need_32bit_api(sbi))
3402 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
3404 stat->ino = inode->i_ino;
3405 stat->mode = inode->i_mode;
3406 stat->uid = inode->i_uid;
3407 stat->gid = inode->i_gid;
3408 stat->rdev = inode->i_rdev;
3409 stat->atime = inode->i_atime;
3410 stat->mtime = inode->i_mtime;
3411 stat->ctime = inode->i_ctime;
3412 stat->blksize = 1 << inode->i_blkbits;
3413 stat->blocks = inode->i_blocks;
3415 if (S_ISDIR(inode->i_mode) &&
3416 ll_i2info(inode)->lli_lsm_md != NULL) {
3417 stat->nlink = lli->lli_stripe_dir_nlink;
3418 stat->size = lli->lli_stripe_dir_size;
3420 stat->nlink = inode->i_nlink;
3421 stat->size = i_size_read(inode);
3427 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3428 __u64 start, __u64 len)
3432 struct ll_user_fiemap *fiemap;
3433 unsigned int extent_count = fieinfo->fi_extents_max;
3435 num_bytes = sizeof(*fiemap) + (extent_count *
3436 sizeof(struct ll_fiemap_extent));
3437 OBD_ALLOC_LARGE(fiemap, num_bytes);
3442 fiemap->fm_flags = fieinfo->fi_flags;
3443 fiemap->fm_extent_count = fieinfo->fi_extents_max;
3444 fiemap->fm_start = start;
3445 fiemap->fm_length = len;
3446 if (extent_count > 0)
3447 memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
3448 sizeof(struct ll_fiemap_extent));
3450 rc = ll_do_fiemap(inode, fiemap, num_bytes);
3452 fieinfo->fi_flags = fiemap->fm_flags;
3453 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
3454 if (extent_count > 0)
3455 memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
3456 fiemap->fm_mapped_extents *
3457 sizeof(struct ll_fiemap_extent));
3459 OBD_FREE_LARGE(fiemap, num_bytes);
3463 struct posix_acl *ll_get_acl(struct inode *inode, int type)
3465 struct ll_inode_info *lli = ll_i2info(inode);
3466 struct posix_acl *acl = NULL;
3469 spin_lock(&lli->lli_lock);
3470 /* VFS' acl_permission_check->check_acl will release the refcount */
3471 acl = posix_acl_dup(lli->lli_posix_acl);
3472 spin_unlock(&lli->lli_lock);
3477 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
3479 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3480 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
3482 ll_check_acl(struct inode *inode, int mask)
3485 # ifdef CONFIG_FS_POSIX_ACL
3486 struct posix_acl *acl;
3490 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
3491 if (flags & IPERM_FLAG_RCU)
3494 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
3499 rc = posix_acl_permission(inode, acl, mask);
3500 posix_acl_release(acl);
3503 # else /* !CONFIG_FS_POSIX_ACL */
3505 # endif /* CONFIG_FS_POSIX_ACL */
3507 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
3509 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
3510 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
3512 # ifdef HAVE_INODE_PERMISION_2ARGS
3513 int ll_inode_permission(struct inode *inode, int mask)
3515 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
3520 struct ll_sb_info *sbi;
3521 struct root_squash_info *squash;
3522 struct cred *cred = NULL;
3523 const struct cred *old_cred = NULL;
3525 bool squash_id = false;
3528 #ifdef MAY_NOT_BLOCK
3529 if (mask & MAY_NOT_BLOCK)
3531 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
3532 if (flags & IPERM_FLAG_RCU)
3536 /* as root inode are NOT getting validated in lookup operation,
3537 * need to do it before permission check. */
3539 if (inode == inode->i_sb->s_root->d_inode) {
3540 rc = __ll_inode_revalidate(inode->i_sb->s_root,
3541 MDS_INODELOCK_LOOKUP);
3546 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
3547 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
3549 /* squash fsuid/fsgid if needed */
3550 sbi = ll_i2sbi(inode);
3551 squash = &sbi->ll_squash;
3552 if (unlikely(squash->rsi_uid != 0 &&
3553 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
3554 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
3558 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
3559 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
3560 squash->rsi_uid, squash->rsi_gid);
3562 /* update current process's credentials
3563 * and FS capability */
3564 cred = prepare_creds();
3568 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
3569 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
3570 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
3571 if ((1 << cap) & CFS_CAP_FS_MASK)
3572 cap_lower(cred->cap_effective, cap);
3574 old_cred = override_creds(cred);
3577 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
3579 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
3580 rc = lustre_check_remote_perm(inode, mask);
3582 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
3584 /* restore current process's credentials and FS capability */
3586 revert_creds(old_cred);
3593 /* -o localflock - only provides locally consistent flock locks */
3594 struct file_operations ll_file_operations = {
3595 .read = ll_file_read,
3596 .aio_read = ll_file_aio_read,
3597 .write = ll_file_write,
3598 .aio_write = ll_file_aio_write,
3599 .unlocked_ioctl = ll_file_ioctl,
3600 .open = ll_file_open,
3601 .release = ll_file_release,
3602 .mmap = ll_file_mmap,
3603 .llseek = ll_file_seek,
3604 .splice_read = ll_file_splice_read,
3609 struct file_operations ll_file_operations_flock = {
3610 .read = ll_file_read,
3611 .aio_read = ll_file_aio_read,
3612 .write = ll_file_write,
3613 .aio_write = ll_file_aio_write,
3614 .unlocked_ioctl = ll_file_ioctl,
3615 .open = ll_file_open,
3616 .release = ll_file_release,
3617 .mmap = ll_file_mmap,
3618 .llseek = ll_file_seek,
3619 .splice_read = ll_file_splice_read,
3622 .flock = ll_file_flock,
3623 .lock = ll_file_flock
3626 /* These are for -o noflock - to return ENOSYS on flock calls */
3627 struct file_operations ll_file_operations_noflock = {
3628 .read = ll_file_read,
3629 .aio_read = ll_file_aio_read,
3630 .write = ll_file_write,
3631 .aio_write = ll_file_aio_write,
3632 .unlocked_ioctl = ll_file_ioctl,
3633 .open = ll_file_open,
3634 .release = ll_file_release,
3635 .mmap = ll_file_mmap,
3636 .llseek = ll_file_seek,
3637 .splice_read = ll_file_splice_read,
3640 .flock = ll_file_noflock,
3641 .lock = ll_file_noflock
3644 struct inode_operations ll_file_inode_operations = {
3645 .setattr = ll_setattr,
3646 .getattr = ll_getattr,
3647 .permission = ll_inode_permission,
3648 .setxattr = ll_setxattr,
3649 .getxattr = ll_getxattr,
3650 .listxattr = ll_listxattr,
3651 .removexattr = ll_removexattr,
3652 .fiemap = ll_fiemap,
3653 #ifdef HAVE_IOP_GET_ACL
3654 .get_acl = ll_get_acl,
3658 /* dynamic ioctl number support routins */
3659 static struct llioc_ctl_data {
3660 struct rw_semaphore ioc_sem;
3661 struct list_head ioc_head;
3663 __RWSEM_INITIALIZER(llioc.ioc_sem),
3664 LIST_HEAD_INIT(llioc.ioc_head)
3669 struct list_head iocd_list;
3670 unsigned int iocd_size;
3671 llioc_callback_t iocd_cb;
3672 unsigned int iocd_count;
3673 unsigned int iocd_cmd[0];
3676 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
3679 struct llioc_data *in_data = NULL;
3682 if (cb == NULL || cmd == NULL ||
3683 count > LLIOC_MAX_CMD || count < 0)
3686 size = sizeof(*in_data) + count * sizeof(unsigned int);
3687 OBD_ALLOC(in_data, size);
3688 if (in_data == NULL)
3691 memset(in_data, 0, sizeof(*in_data));
3692 in_data->iocd_size = size;
3693 in_data->iocd_cb = cb;
3694 in_data->iocd_count = count;
3695 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
3697 down_write(&llioc.ioc_sem);
3698 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
3699 up_write(&llioc.ioc_sem);
3704 void ll_iocontrol_unregister(void *magic)
3706 struct llioc_data *tmp;
3711 down_write(&llioc.ioc_sem);
3712 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
3714 unsigned int size = tmp->iocd_size;
3716 list_del(&tmp->iocd_list);
3717 up_write(&llioc.ioc_sem);
3719 OBD_FREE(tmp, size);
3723 up_write(&llioc.ioc_sem);
3725 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
3728 EXPORT_SYMBOL(ll_iocontrol_register);
3729 EXPORT_SYMBOL(ll_iocontrol_unregister);
3731 static enum llioc_iter
3732 ll_iocontrol_call(struct inode *inode, struct file *file,
3733 unsigned int cmd, unsigned long arg, int *rcp)
3735 enum llioc_iter ret = LLIOC_CONT;
3736 struct llioc_data *data;
3737 int rc = -EINVAL, i;
3739 down_read(&llioc.ioc_sem);
3740 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
3741 for (i = 0; i < data->iocd_count; i++) {
3742 if (cmd != data->iocd_cmd[i])
3745 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
3749 if (ret == LLIOC_STOP)
3752 up_read(&llioc.ioc_sem);
3759 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
3761 struct ll_inode_info *lli = ll_i2info(inode);
3762 struct cl_env_nest nest;
3767 if (lli->lli_clob == NULL)
3770 env = cl_env_nested_get(&nest);
3772 RETURN(PTR_ERR(env));
3774 result = cl_conf_set(env, lli->lli_clob, conf);
3775 cl_env_nested_put(&nest, env);
3777 if (conf->coc_opc == OBJECT_CONF_SET) {
3778 struct ldlm_lock *lock = conf->coc_lock;
3780 LASSERT(lock != NULL);
3781 LASSERT(ldlm_has_layout(lock));
3783 struct lustre_md *md = conf->u.coc_md;
3784 __u32 gen = LL_LAYOUT_GEN_EMPTY;
3786 /* it can only be allowed to match after layout is
3787 * applied to inode otherwise false layout would be
3788 * seen. Applying layout shoud happen before dropping
3789 * the intent lock. */
3790 ldlm_lock_allow_match(lock);
3792 lli->lli_has_smd = lsm_has_objects(md->lsm);
3793 if (md->lsm != NULL)
3794 gen = md->lsm->lsm_layout_gen;
3797 DFID ": layout version change: %u -> %u\n",
3798 PFID(&lli->lli_fid), ll_layout_version_get(lli),
3800 ll_layout_version_set(lli, gen);
3806 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
3807 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
3810 struct ll_sb_info *sbi = ll_i2sbi(inode);
3811 struct obd_capa *oc;
3812 struct ptlrpc_request *req;
3813 struct mdt_body *body;
3820 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
3821 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
3822 lock->l_lvb_data, lock->l_lvb_len);
3824 if ((lock->l_lvb_data != NULL) && ldlm_is_lvb_ready(lock))
3827 /* if layout lock was granted right away, the layout is returned
3828 * within DLM_LVB of dlm reply; otherwise if the lock was ever
3829 * blocked and then granted via completion ast, we have to fetch
3830 * layout here. Please note that we can't use the LVB buffer in
3831 * completion AST because it doesn't have a large enough buffer */
3832 oc = ll_mdscapa_get(inode);
3833 rc = ll_get_default_mdsize(sbi, &lmmsize);
3835 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
3836 OBD_MD_FLXATTR, XATTR_NAME_LOV, NULL, 0,
3842 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3844 GOTO(out, rc = -EPROTO);
3846 lmmsize = body->mbo_eadatasize;
3847 if (lmmsize == 0) /* empty layout */
3850 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
3852 GOTO(out, rc = -EFAULT);
3854 OBD_ALLOC_LARGE(lvbdata, lmmsize);
3855 if (lvbdata == NULL)
3856 GOTO(out, rc = -ENOMEM);
3858 memcpy(lvbdata, lmm, lmmsize);
3859 lock_res_and_lock(lock);
3860 if (lock->l_lvb_data != NULL)
3861 OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len);
3863 lock->l_lvb_data = lvbdata;
3864 lock->l_lvb_len = lmmsize;
3865 unlock_res_and_lock(lock);
3870 ptlrpc_req_finished(req);
3875 * Apply the layout to the inode. Layout lock is held and will be released
3878 static int ll_layout_lock_set(struct lustre_handle *lockh, ldlm_mode_t mode,
3879 struct inode *inode, __u32 *gen, bool reconf)
3881 struct ll_inode_info *lli = ll_i2info(inode);
3882 struct ll_sb_info *sbi = ll_i2sbi(inode);
3883 struct ldlm_lock *lock;
3884 struct lustre_md md = { NULL };
3885 struct cl_object_conf conf;
3888 bool wait_layout = false;
3891 LASSERT(lustre_handle_is_used(lockh));
3893 lock = ldlm_handle2lock(lockh);
3894 LASSERT(lock != NULL);
3895 LASSERT(ldlm_has_layout(lock));
3897 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured: %d\n",
3898 PFID(&lli->lli_fid), inode, reconf);
3900 /* in case this is a caching lock and reinstate with new inode */
3901 md_set_lock_data(sbi->ll_md_exp, &lockh->cookie, inode, NULL);
3903 lock_res_and_lock(lock);
3904 lvb_ready = ldlm_is_lvb_ready(lock);
3905 unlock_res_and_lock(lock);
3906 /* checking lvb_ready is racy but this is okay. The worst case is
3907 * that multi processes may configure the file on the same time. */
3909 if (lvb_ready || !reconf) {
3912 /* layout_gen must be valid if layout lock is not
3913 * cancelled and stripe has already set */
3914 *gen = ll_layout_version_get(lli);
3920 rc = ll_layout_fetch(inode, lock);
3924 /* for layout lock, lmm is returned in lock's lvb.
3925 * lvb_data is immutable if the lock is held so it's safe to access it
3926 * without res lock. See the description in ldlm_lock_decref_internal()
3927 * for the condition to free lvb_data of layout lock */
3928 if (lock->l_lvb_data != NULL) {
3929 rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm,
3930 lock->l_lvb_data, lock->l_lvb_len);
3932 *gen = LL_LAYOUT_GEN_EMPTY;
3934 *gen = md.lsm->lsm_layout_gen;
3937 CERROR("%s: file "DFID" unpackmd error: %d\n",
3938 ll_get_fsname(inode->i_sb, NULL, 0),
3939 PFID(&lli->lli_fid), rc);
3945 /* set layout to file. Unlikely this will fail as old layout was
3946 * surely eliminated */
3947 memset(&conf, 0, sizeof conf);
3948 conf.coc_opc = OBJECT_CONF_SET;
3949 conf.coc_inode = inode;
3950 conf.coc_lock = lock;
3951 conf.u.coc_md = &md;
3952 rc = ll_layout_conf(inode, &conf);
3955 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
3957 /* refresh layout failed, need to wait */
3958 wait_layout = rc == -EBUSY;
3962 LDLM_LOCK_PUT(lock);
3963 ldlm_lock_decref(lockh, mode);
3965 /* wait for IO to complete if it's still being used. */
3967 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
3968 ll_get_fsname(inode->i_sb, NULL, 0),
3969 PFID(&lli->lli_fid), inode);
3971 memset(&conf, 0, sizeof conf);
3972 conf.coc_opc = OBJECT_CONF_WAIT;
3973 conf.coc_inode = inode;
3974 rc = ll_layout_conf(inode, &conf);
3978 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
3979 ll_get_fsname(inode->i_sb, NULL, 0),
3980 PFID(&lli->lli_fid), rc);
3986 * This function checks if there exists a LAYOUT lock on the client side,
3987 * or enqueues it if it doesn't have one in cache.
3989 * This function will not hold layout lock so it may be revoked any time after
3990 * this function returns. Any operations depend on layout should be redone
3993 * This function should be called before lov_io_init() to get an uptodate
3994 * layout version, the caller should save the version number and after IO
3995 * is finished, this function should be called again to verify that layout
3996 * is not changed during IO time.
3998 int ll_layout_refresh(struct inode *inode, __u32 *gen)
4000 struct ll_inode_info *lli = ll_i2info(inode);
4001 struct ll_sb_info *sbi = ll_i2sbi(inode);
4002 struct md_op_data *op_data;
4003 struct lookup_intent it;
4004 struct lustre_handle lockh;
4006 struct ldlm_enqueue_info einfo = {
4007 .ei_type = LDLM_IBITS,
4009 .ei_cb_bl = &ll_md_blocking_ast,
4010 .ei_cb_cp = &ldlm_completion_ast,
4015 *gen = ll_layout_version_get(lli);
4016 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != LL_LAYOUT_GEN_NONE)
4020 LASSERT(fid_is_sane(ll_inode2fid(inode)));
4021 LASSERT(S_ISREG(inode->i_mode));
4023 /* take layout lock mutex to enqueue layout lock exclusively. */
4024 mutex_lock(&lli->lli_layout_mutex);
4027 /* mostly layout lock is caching on the local side, so try to match
4028 * it before grabbing layout lock mutex. */
4029 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
4030 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
4031 if (mode != 0) { /* hit cached lock */
4032 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
4036 mutex_unlock(&lli->lli_layout_mutex);
4040 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
4041 0, 0, LUSTRE_OPC_ANY, NULL);
4042 if (IS_ERR(op_data)) {
4043 mutex_unlock(&lli->lli_layout_mutex);
4044 RETURN(PTR_ERR(op_data));
4047 /* have to enqueue one */
4048 memset(&it, 0, sizeof(it));
4049 it.it_op = IT_LAYOUT;
4050 lockh.cookie = 0ULL;
4052 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)\n",
4053 ll_get_fsname(inode->i_sb, NULL, 0),
4054 PFID(&lli->lli_fid), inode);
4056 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, &it, op_data, &lockh, 0);
4057 if (it.d.lustre.it_data != NULL)
4058 ptlrpc_req_finished(it.d.lustre.it_data);
4059 it.d.lustre.it_data = NULL;
4061 ll_finish_md_op_data(op_data);
4063 mode = it.d.lustre.it_lock_mode;
4064 it.d.lustre.it_lock_mode = 0;
4065 ll_intent_drop_lock(&it);
4068 /* set lock data in case this is a new lock */
4069 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
4070 rc = ll_layout_lock_set(&lockh, mode, inode, gen, true);
4074 mutex_unlock(&lli->lli_layout_mutex);
4080 * This function send a restore request to the MDT
4082 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
4084 struct hsm_user_request *hur;
4088 len = sizeof(struct hsm_user_request) +
4089 sizeof(struct hsm_user_item);
4090 OBD_ALLOC(hur, len);
4094 hur->hur_request.hr_action = HUA_RESTORE;
4095 hur->hur_request.hr_archive_id = 0;
4096 hur->hur_request.hr_flags = 0;
4097 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
4098 sizeof(hur->hur_user_item[0].hui_fid));
4099 hur->hur_user_item[0].hui_extent.offset = offset;
4100 hur->hur_user_item[0].hui_extent.length = length;
4101 hur->hur_request.hr_itemcount = 1;
4102 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, cl_i2sbi(inode)->ll_md_exp,