4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Phil Schwan <phil@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LLITE
40 #include <lustre_dlm.h>
41 #include <linux/pagemap.h>
42 #include <linux/file.h>
43 #include <linux/sched.h>
44 #include <linux/user_namespace.h>
45 #ifdef HAVE_UIDGID_HEADER
46 # include <linux/uidgid.h>
49 #include <uapi/linux/lustre/lustre_ioctl.h>
50 #include <lustre_swab.h>
52 #include "cl_object.h"
53 #include "llite_internal.h"
54 #include "vvp_internal.h"
57 struct inode *sp_inode;
62 __u64 pa_data_version;
68 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
70 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
73 static struct ll_file_data *ll_file_data_get(void)
75 struct ll_file_data *fd;
77 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
81 fd->fd_write_failed = false;
82 pcc_file_init(&fd->fd_pcc_file);
87 static void ll_file_data_put(struct ll_file_data *fd)
90 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
94 * Packs all the attributes into @op_data for the CLOSE rpc.
96 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
97 struct obd_client_handle *och)
101 ll_prep_md_op_data(op_data, inode, NULL, NULL,
102 0, 0, LUSTRE_OPC_ANY, NULL);
104 op_data->op_attr.ia_mode = inode->i_mode;
105 op_data->op_attr.ia_atime = inode->i_atime;
106 op_data->op_attr.ia_mtime = inode->i_mtime;
107 op_data->op_attr.ia_ctime = inode->i_ctime;
108 op_data->op_attr.ia_size = i_size_read(inode);
109 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
110 ATTR_MTIME | ATTR_MTIME_SET |
112 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
113 op_data->op_attr_blocks = inode->i_blocks;
114 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
115 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
116 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
117 op_data->op_open_handle = och->och_open_handle;
119 if (och->och_flags & FMODE_WRITE &&
120 ll_file_test_and_clear_flag(ll_i2info(inode), LLIF_DATA_MODIFIED))
121 /* For HSM: if inode data has been modified, pack it so that
122 * MDT can set data dirty flag in the archive. */
123 op_data->op_bias |= MDS_DATA_MODIFIED;
129 * Perform a close, possibly with a bias.
130 * The meaning of "data" depends on the value of "bias".
132 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
133 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
136 static int ll_close_inode_openhandle(struct inode *inode,
137 struct obd_client_handle *och,
138 enum mds_op_bias bias, void *data)
140 struct obd_export *md_exp = ll_i2mdexp(inode);
141 const struct ll_inode_info *lli = ll_i2info(inode);
142 struct md_op_data *op_data;
143 struct ptlrpc_request *req = NULL;
147 if (class_exp2obd(md_exp) == NULL) {
148 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
149 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
153 OBD_ALLOC_PTR(op_data);
154 /* We leak openhandle and request here on error, but not much to be
155 * done in OOM case since app won't retry close on error either. */
157 GOTO(out, rc = -ENOMEM);
159 ll_prepare_close(inode, op_data, och);
161 case MDS_CLOSE_LAYOUT_MERGE:
162 /* merge blocks from the victim inode */
163 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
164 op_data->op_attr.ia_valid |= ATTR_SIZE;
165 op_data->op_xvalid |= OP_XVALID_BLOCKS;
166 case MDS_CLOSE_LAYOUT_SPLIT:
167 case MDS_CLOSE_LAYOUT_SWAP: {
168 struct split_param *sp = data;
170 LASSERT(data != NULL);
171 op_data->op_bias |= bias;
172 op_data->op_data_version = 0;
173 op_data->op_lease_handle = och->och_lease_handle;
174 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
175 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
176 op_data->op_mirror_id = sp->sp_mirror_id;
178 op_data->op_fid2 = *ll_inode2fid(data);
183 case MDS_CLOSE_RESYNC_DONE: {
184 struct ll_ioc_lease *ioc = data;
186 LASSERT(data != NULL);
187 op_data->op_attr_blocks +=
188 ioc->lil_count * op_data->op_attr_blocks;
189 op_data->op_attr.ia_valid |= ATTR_SIZE;
190 op_data->op_xvalid |= OP_XVALID_BLOCKS;
191 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
193 op_data->op_lease_handle = och->och_lease_handle;
194 op_data->op_data = &ioc->lil_ids[0];
195 op_data->op_data_size =
196 ioc->lil_count * sizeof(ioc->lil_ids[0]);
200 case MDS_PCC_ATTACH: {
201 struct pcc_param *param = data;
203 LASSERT(data != NULL);
204 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
205 op_data->op_archive_id = param->pa_archive_id;
206 op_data->op_data_version = param->pa_data_version;
207 op_data->op_lease_handle = och->och_lease_handle;
211 case MDS_HSM_RELEASE:
212 LASSERT(data != NULL);
213 op_data->op_bias |= MDS_HSM_RELEASE;
214 op_data->op_data_version = *(__u64 *)data;
215 op_data->op_lease_handle = och->och_lease_handle;
216 op_data->op_attr.ia_valid |= ATTR_SIZE;
217 op_data->op_xvalid |= OP_XVALID_BLOCKS;
221 LASSERT(data == NULL);
225 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
226 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
227 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
228 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
230 rc = md_close(md_exp, op_data, och->och_mod, &req);
231 if (rc != 0 && rc != -EINTR)
232 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
233 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
235 if (rc == 0 && op_data->op_bias & bias) {
236 struct mdt_body *body;
238 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
239 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
242 if (bias & MDS_PCC_ATTACH) {
243 struct pcc_param *param = data;
245 param->pa_layout_gen = body->mbo_layout_gen;
249 ll_finish_md_op_data(op_data);
253 md_clear_open_replay_data(md_exp, och);
254 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
257 ptlrpc_req_finished(req); /* This is close request */
261 int ll_md_real_close(struct inode *inode, fmode_t fmode)
263 struct ll_inode_info *lli = ll_i2info(inode);
264 struct obd_client_handle **och_p;
265 struct obd_client_handle *och;
270 if (fmode & FMODE_WRITE) {
271 och_p = &lli->lli_mds_write_och;
272 och_usecount = &lli->lli_open_fd_write_count;
273 } else if (fmode & FMODE_EXEC) {
274 och_p = &lli->lli_mds_exec_och;
275 och_usecount = &lli->lli_open_fd_exec_count;
277 LASSERT(fmode & FMODE_READ);
278 och_p = &lli->lli_mds_read_och;
279 och_usecount = &lli->lli_open_fd_read_count;
282 mutex_lock(&lli->lli_och_mutex);
283 if (*och_usecount > 0) {
284 /* There are still users of this handle, so skip
286 mutex_unlock(&lli->lli_och_mutex);
292 mutex_unlock(&lli->lli_och_mutex);
295 /* There might be a race and this handle may already
297 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
303 static int ll_md_close(struct inode *inode, struct file *file)
305 union ldlm_policy_data policy = {
306 .l_inodebits = { MDS_INODELOCK_OPEN },
308 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
309 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
310 struct ll_inode_info *lli = ll_i2info(inode);
311 struct lustre_handle lockh;
312 enum ldlm_mode lockmode;
316 /* clear group lock, if present */
317 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
318 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
320 if (fd->fd_lease_och != NULL) {
323 /* Usually the lease is not released when the
324 * application crashed, we need to release here. */
325 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
326 CDEBUG(rc ? D_ERROR : D_INODE, "Clean up lease "DFID" %d/%d\n",
327 PFID(&lli->lli_fid), rc, lease_broken);
329 fd->fd_lease_och = NULL;
332 if (fd->fd_och != NULL) {
333 rc = ll_close_inode_openhandle(inode, fd->fd_och, 0, NULL);
338 /* Let's see if we have good enough OPEN lock on the file and if
339 we can skip talking to MDS */
340 mutex_lock(&lli->lli_och_mutex);
341 if (fd->fd_omode & FMODE_WRITE) {
343 LASSERT(lli->lli_open_fd_write_count);
344 lli->lli_open_fd_write_count--;
345 } else if (fd->fd_omode & FMODE_EXEC) {
347 LASSERT(lli->lli_open_fd_exec_count);
348 lli->lli_open_fd_exec_count--;
351 LASSERT(lli->lli_open_fd_read_count);
352 lli->lli_open_fd_read_count--;
354 mutex_unlock(&lli->lli_och_mutex);
356 if (!md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
357 LDLM_IBITS, &policy, lockmode, &lockh))
358 rc = ll_md_real_close(inode, fd->fd_omode);
361 LUSTRE_FPRIVATE(file) = NULL;
362 ll_file_data_put(fd);
367 /* While this returns an error code, fput() the caller does not, so we need
368 * to make every effort to clean up all of our state here. Also, applications
369 * rarely check close errors and even if an error is returned they will not
370 * re-try the close call.
372 int ll_file_release(struct inode *inode, struct file *file)
374 struct ll_file_data *fd;
375 struct ll_sb_info *sbi = ll_i2sbi(inode);
376 struct ll_inode_info *lli = ll_i2info(inode);
380 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
381 PFID(ll_inode2fid(inode)), inode);
383 if (inode->i_sb->s_root != file_dentry(file))
384 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
385 fd = LUSTRE_FPRIVATE(file);
388 /* The last ref on @file, maybe not the the owner pid of statahead,
389 * because parent and child process can share the same file handle. */
390 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
391 ll_deauthorize_statahead(inode, fd);
393 if (inode->i_sb->s_root == file_dentry(file)) {
394 LUSTRE_FPRIVATE(file) = NULL;
395 ll_file_data_put(fd);
399 pcc_file_release(inode, file);
401 if (!S_ISDIR(inode->i_mode)) {
402 if (lli->lli_clob != NULL)
403 lov_read_and_clear_async_rc(lli->lli_clob);
404 lli->lli_async_rc = 0;
407 rc = ll_md_close(inode, file);
409 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
410 libcfs_debug_dumplog();
415 static inline int ll_dom_readpage(void *data, struct page *page)
417 struct niobuf_local *lnb = data;
420 kaddr = ll_kmap_atomic(page, KM_USER0);
421 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
422 if (lnb->lnb_len < PAGE_SIZE)
423 memset(kaddr + lnb->lnb_len, 0,
424 PAGE_SIZE - lnb->lnb_len);
425 flush_dcache_page(page);
426 SetPageUptodate(page);
427 ll_kunmap_atomic(kaddr, KM_USER0);
433 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req,
434 struct lookup_intent *it)
436 struct ll_inode_info *lli = ll_i2info(inode);
437 struct cl_object *obj = lli->lli_clob;
438 struct address_space *mapping = inode->i_mapping;
440 struct niobuf_remote *rnb;
441 struct mdt_body *body;
443 unsigned long index, start;
444 struct niobuf_local lnb;
451 if (!req_capsule_has_field(&req->rq_pill, &RMF_NIOBUF_INLINE,
455 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
456 if (rnb == NULL || rnb->rnb_len == 0)
459 /* LU-11595: Server may return whole file and that is OK always or
460 * it may return just file tail and its offset must be aligned with
461 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
462 * smaller then offset may be not aligned and that data is just ignored.
464 if (rnb->rnb_offset % PAGE_SIZE)
467 /* Server returns whole file or just file tail if it fills in reply
468 * buffer, in both cases total size should be equal to the file size.
470 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
471 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size) {
472 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
473 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
474 rnb->rnb_len, body->mbo_dom_size);
478 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
479 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
481 data = (char *)rnb + sizeof(*rnb);
483 lnb.lnb_file_offset = rnb->rnb_offset;
484 start = lnb.lnb_file_offset / PAGE_SIZE;
486 LASSERT(lnb.lnb_file_offset % PAGE_SIZE == 0);
487 lnb.lnb_page_offset = 0;
489 lnb.lnb_data = data + (index << PAGE_SHIFT);
490 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
491 if (lnb.lnb_len > PAGE_SIZE)
492 lnb.lnb_len = PAGE_SIZE;
494 vmpage = read_cache_page(mapping, index + start,
495 ll_dom_readpage, &lnb);
496 if (IS_ERR(vmpage)) {
497 CWARN("%s: cannot fill page %lu for "DFID
498 " with data: rc = %li\n",
499 ll_i2sbi(inode)->ll_fsname, index + start,
500 PFID(lu_object_fid(&obj->co_lu)),
506 } while (rnb->rnb_len > (index << PAGE_SHIFT));
510 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
511 struct lookup_intent *itp)
513 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
514 struct dentry *parent = de->d_parent;
517 struct md_op_data *op_data;
518 struct ptlrpc_request *req = NULL;
522 LASSERT(parent != NULL);
523 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
525 /* if server supports open-by-fid, or file name is invalid, don't pack
526 * name in open request */
527 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
528 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
530 len = de->d_name.len;
531 name = kmalloc(len + 1, GFP_NOFS);
536 spin_lock(&de->d_lock);
537 if (len != de->d_name.len) {
538 spin_unlock(&de->d_lock);
542 memcpy(name, de->d_name.name, len);
544 spin_unlock(&de->d_lock);
546 if (!lu_name_is_valid_2(name, len)) {
552 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
553 name, len, 0, LUSTRE_OPC_ANY, NULL);
554 if (IS_ERR(op_data)) {
556 RETURN(PTR_ERR(op_data));
558 op_data->op_data = lmm;
559 op_data->op_data_size = lmmsize;
561 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
562 &ll_md_blocking_ast, 0);
564 ll_finish_md_op_data(op_data);
566 /* reason for keep own exit path - don`t flood log
567 * with messages with -ESTALE errors.
569 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
570 it_open_error(DISP_OPEN_OPEN, itp))
572 ll_release_openhandle(de, itp);
576 if (it_disposition(itp, DISP_LOOKUP_NEG))
577 GOTO(out, rc = -ENOENT);
579 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
580 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
581 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
585 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
587 if (!rc && itp->it_lock_mode) {
588 struct lustre_handle handle = {.cookie = itp->it_lock_handle};
589 struct ldlm_lock *lock;
590 bool has_dom_bit = false;
592 /* If we got a lock back and it has a LOOKUP bit set,
593 * make sure the dentry is marked as valid so we can find it.
594 * We don't need to care about actual hashing since other bits
595 * of kernel will deal with that later.
597 lock = ldlm_handle2lock(&handle);
599 has_dom_bit = ldlm_has_dom(lock);
600 if (lock->l_policy_data.l_inodebits.bits &
601 MDS_INODELOCK_LOOKUP)
602 d_lustre_revalidate(de);
606 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
608 ll_dom_finish_open(de->d_inode, req, itp);
612 ptlrpc_req_finished(req);
613 ll_intent_drop_lock(itp);
615 /* We did open by fid, but by the time we got to the server,
616 * the object disappeared. If this is a create, we cannot really
617 * tell the userspace that the file it was trying to create
618 * does not exist. Instead let's return -ESTALE, and the VFS will
619 * retry the create with LOOKUP_REVAL that we are going to catch
620 * in ll_revalidate_dentry() and use lookup then.
622 if (rc == -ENOENT && itp->it_op & IT_CREAT)
628 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
629 struct obd_client_handle *och)
631 struct mdt_body *body;
633 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
634 och->och_open_handle = body->mbo_open_handle;
635 och->och_fid = body->mbo_fid1;
636 och->och_lease_handle.cookie = it->it_lock_handle;
637 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
638 och->och_flags = it->it_flags;
640 return md_set_open_replay_data(md_exp, och, it);
643 static int ll_local_open(struct file *file, struct lookup_intent *it,
644 struct ll_file_data *fd, struct obd_client_handle *och)
646 struct inode *inode = file_inode(file);
649 LASSERT(!LUSTRE_FPRIVATE(file));
656 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
661 LUSTRE_FPRIVATE(file) = fd;
662 ll_readahead_init(inode, &fd->fd_ras);
663 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
665 /* ll_cl_context initialize */
666 rwlock_init(&fd->fd_lock);
667 INIT_LIST_HEAD(&fd->fd_lccs);
672 /* Open a file, and (for the very first open) create objects on the OSTs at
673 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
674 * creation or open until ll_lov_setstripe() ioctl is called.
676 * If we already have the stripe MD locally then we don't request it in
677 * md_open(), by passing a lmm_size = 0.
679 * It is up to the application to ensure no other processes open this file
680 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
681 * used. We might be able to avoid races of that sort by getting lli_open_sem
682 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
683 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
685 int ll_file_open(struct inode *inode, struct file *file)
687 struct ll_inode_info *lli = ll_i2info(inode);
688 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
689 .it_flags = file->f_flags };
690 struct obd_client_handle **och_p = NULL;
691 __u64 *och_usecount = NULL;
692 struct ll_file_data *fd;
696 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
697 PFID(ll_inode2fid(inode)), inode, file->f_flags);
699 it = file->private_data; /* XXX: compat macro */
700 file->private_data = NULL; /* prevent ll_local_open assertion */
702 fd = ll_file_data_get();
704 GOTO(out_nofiledata, rc = -ENOMEM);
707 if (S_ISDIR(inode->i_mode))
708 ll_authorize_statahead(inode, fd);
710 if (inode->i_sb->s_root == file_dentry(file)) {
711 LUSTRE_FPRIVATE(file) = fd;
715 if (!it || !it->it_disposition) {
716 /* Convert f_flags into access mode. We cannot use file->f_mode,
717 * because everything but O_ACCMODE mask was stripped from
719 if ((oit.it_flags + 1) & O_ACCMODE)
721 if (file->f_flags & O_TRUNC)
722 oit.it_flags |= FMODE_WRITE;
724 /* kernel only call f_op->open in dentry_open. filp_open calls
725 * dentry_open after call to open_namei that checks permissions.
726 * Only nfsd_open call dentry_open directly without checking
727 * permissions and because of that this code below is safe.
729 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
730 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
732 /* We do not want O_EXCL here, presumably we opened the file
733 * already? XXX - NFS implications? */
734 oit.it_flags &= ~O_EXCL;
736 /* bug20584, if "it_flags" contains O_CREAT, the file will be
737 * created if necessary, then "IT_CREAT" should be set to keep
738 * consistent with it */
739 if (oit.it_flags & O_CREAT)
740 oit.it_op |= IT_CREAT;
746 /* Let's see if we have file open on MDS already. */
747 if (it->it_flags & FMODE_WRITE) {
748 och_p = &lli->lli_mds_write_och;
749 och_usecount = &lli->lli_open_fd_write_count;
750 } else if (it->it_flags & FMODE_EXEC) {
751 och_p = &lli->lli_mds_exec_och;
752 och_usecount = &lli->lli_open_fd_exec_count;
754 och_p = &lli->lli_mds_read_och;
755 och_usecount = &lli->lli_open_fd_read_count;
758 mutex_lock(&lli->lli_och_mutex);
759 if (*och_p) { /* Open handle is present */
760 if (it_disposition(it, DISP_OPEN_OPEN)) {
761 /* Well, there's extra open request that we do not need,
762 let's close it somehow. This will decref request. */
763 rc = it_open_error(DISP_OPEN_OPEN, it);
765 mutex_unlock(&lli->lli_och_mutex);
766 GOTO(out_openerr, rc);
769 ll_release_openhandle(file_dentry(file), it);
773 rc = ll_local_open(file, it, fd, NULL);
776 mutex_unlock(&lli->lli_och_mutex);
777 GOTO(out_openerr, rc);
780 LASSERT(*och_usecount == 0);
781 if (!it->it_disposition) {
782 struct ll_dentry_data *ldd = ll_d2d(file->f_path.dentry);
783 /* We cannot just request lock handle now, new ELC code
784 means that one of other OPEN locks for this file
785 could be cancelled, and since blocking ast handler
786 would attempt to grab och_mutex as well, that would
787 result in a deadlock */
788 mutex_unlock(&lli->lli_och_mutex);
790 * Normally called under two situations:
792 * 2. A race/condition on MDS resulting in no open
793 * handle to be returned from LOOKUP|OPEN request,
794 * for example if the target entry was a symlink.
796 * Only fetch MDS_OPEN_LOCK if this is in NFS path,
797 * marked by a bit set in ll_iget_for_nfs. Clear the
798 * bit so that it's not confusing later callers.
800 * NB; when ldd is NULL, it must have come via normal
801 * lookup path only, since ll_iget_for_nfs always calls
804 if (ldd && ldd->lld_nfs_dentry) {
805 ldd->lld_nfs_dentry = 0;
806 it->it_flags |= MDS_OPEN_LOCK;
810 * Always specify MDS_OPEN_BY_FID because we don't want
811 * to get file with different fid.
813 it->it_flags |= MDS_OPEN_BY_FID;
814 rc = ll_intent_file_open(file_dentry(file), NULL, 0,
817 GOTO(out_openerr, rc);
821 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
823 GOTO(out_och_free, rc = -ENOMEM);
827 /* md_intent_lock() didn't get a request ref if there was an
828 * open error, so don't do cleanup on the request here
830 /* XXX (green): Should not we bail out on any error here, not
831 * just open error? */
832 rc = it_open_error(DISP_OPEN_OPEN, it);
834 GOTO(out_och_free, rc);
836 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
837 "inode %p: disposition %x, status %d\n", inode,
838 it_disposition(it, ~0), it->it_status);
840 rc = ll_local_open(file, it, fd, *och_p);
842 GOTO(out_och_free, rc);
845 rc = pcc_file_open(inode, file);
847 GOTO(out_och_free, rc);
849 mutex_unlock(&lli->lli_och_mutex);
852 /* Must do this outside lli_och_mutex lock to prevent deadlock where
853 different kind of OPEN lock for this same inode gets cancelled
854 by ldlm_cancel_lru */
855 if (!S_ISREG(inode->i_mode))
856 GOTO(out_och_free, rc);
858 cl_lov_delay_create_clear(&file->f_flags);
859 GOTO(out_och_free, rc);
863 if (och_p && *och_p) {
864 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
865 *och_p = NULL; /* OBD_FREE writes some magic there */
868 mutex_unlock(&lli->lli_och_mutex);
871 if (lli->lli_opendir_key == fd)
872 ll_deauthorize_statahead(inode, fd);
875 ll_file_data_put(fd);
877 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
881 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
882 ptlrpc_req_finished(it->it_request);
883 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
889 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
890 struct ldlm_lock_desc *desc, void *data, int flag)
893 struct lustre_handle lockh;
897 case LDLM_CB_BLOCKING:
898 ldlm_lock2handle(lock, &lockh);
899 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
901 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
905 case LDLM_CB_CANCELING:
913 * When setting a lease on a file, we take ownership of the lli_mds_*_och
914 * and save it as fd->fd_och so as to force client to reopen the file even
915 * if it has an open lock in cache already.
917 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
918 struct lustre_handle *old_open_handle)
920 struct ll_inode_info *lli = ll_i2info(inode);
921 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
922 struct obd_client_handle **och_p;
927 /* Get the openhandle of the file */
928 mutex_lock(&lli->lli_och_mutex);
929 if (fd->fd_lease_och != NULL)
930 GOTO(out_unlock, rc = -EBUSY);
932 if (fd->fd_och == NULL) {
933 if (file->f_mode & FMODE_WRITE) {
934 LASSERT(lli->lli_mds_write_och != NULL);
935 och_p = &lli->lli_mds_write_och;
936 och_usecount = &lli->lli_open_fd_write_count;
938 LASSERT(lli->lli_mds_read_och != NULL);
939 och_p = &lli->lli_mds_read_och;
940 och_usecount = &lli->lli_open_fd_read_count;
943 if (*och_usecount > 1)
944 GOTO(out_unlock, rc = -EBUSY);
951 *old_open_handle = fd->fd_och->och_open_handle;
955 mutex_unlock(&lli->lli_och_mutex);
960 * Release ownership on lli_mds_*_och when putting back a file lease.
962 static int ll_lease_och_release(struct inode *inode, struct file *file)
964 struct ll_inode_info *lli = ll_i2info(inode);
965 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
966 struct obd_client_handle **och_p;
967 struct obd_client_handle *old_och = NULL;
972 mutex_lock(&lli->lli_och_mutex);
973 if (file->f_mode & FMODE_WRITE) {
974 och_p = &lli->lli_mds_write_och;
975 och_usecount = &lli->lli_open_fd_write_count;
977 och_p = &lli->lli_mds_read_och;
978 och_usecount = &lli->lli_open_fd_read_count;
981 /* The file may have been open by another process (broken lease) so
982 * *och_p is not NULL. In this case we should simply increase usecount
985 if (*och_p != NULL) {
986 old_och = fd->fd_och;
993 mutex_unlock(&lli->lli_och_mutex);
996 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1002 * Acquire a lease and open the file.
1004 static struct obd_client_handle *
1005 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1008 struct lookup_intent it = { .it_op = IT_OPEN };
1009 struct ll_sb_info *sbi = ll_i2sbi(inode);
1010 struct md_op_data *op_data;
1011 struct ptlrpc_request *req = NULL;
1012 struct lustre_handle old_open_handle = { 0 };
1013 struct obd_client_handle *och = NULL;
1018 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1019 RETURN(ERR_PTR(-EINVAL));
1022 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1023 RETURN(ERR_PTR(-EPERM));
1025 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1027 RETURN(ERR_PTR(rc));
1032 RETURN(ERR_PTR(-ENOMEM));
1034 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1035 LUSTRE_OPC_ANY, NULL);
1036 if (IS_ERR(op_data))
1037 GOTO(out, rc = PTR_ERR(op_data));
1039 /* To tell the MDT this openhandle is from the same owner */
1040 op_data->op_open_handle = old_open_handle;
1042 it.it_flags = fmode | open_flags;
1043 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1044 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1045 &ll_md_blocking_lease_ast,
1046 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1047 * it can be cancelled which may mislead applications that the lease is
1049 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1050 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1051 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1052 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1053 ll_finish_md_op_data(op_data);
1054 ptlrpc_req_finished(req);
1056 GOTO(out_release_it, rc);
1058 if (it_disposition(&it, DISP_LOOKUP_NEG))
1059 GOTO(out_release_it, rc = -ENOENT);
1061 rc = it_open_error(DISP_OPEN_OPEN, &it);
1063 GOTO(out_release_it, rc);
1065 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1066 ll_och_fill(sbi->ll_md_exp, &it, och);
1068 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1069 GOTO(out_close, rc = -EOPNOTSUPP);
1071 /* already get lease, handle lease lock */
1072 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1073 if (it.it_lock_mode == 0 ||
1074 it.it_lock_bits != MDS_INODELOCK_OPEN) {
1075 /* open lock must return for lease */
1076 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1077 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1079 GOTO(out_close, rc = -EPROTO);
1082 ll_intent_release(&it);
1086 /* Cancel open lock */
1087 if (it.it_lock_mode != 0) {
1088 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1090 it.it_lock_mode = 0;
1091 och->och_lease_handle.cookie = 0ULL;
1093 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1095 CERROR("%s: error closing file "DFID": %d\n",
1096 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1097 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1099 ll_intent_release(&it);
1103 RETURN(ERR_PTR(rc));
1107 * Check whether a layout swap can be done between two inodes.
1109 * \param[in] inode1 First inode to check
1110 * \param[in] inode2 Second inode to check
1112 * \retval 0 on success, layout swap can be performed between both inodes
1113 * \retval negative error code if requirements are not met
1115 static int ll_check_swap_layouts_validity(struct inode *inode1,
1116 struct inode *inode2)
1118 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1121 if (inode_permission(inode1, MAY_WRITE) ||
1122 inode_permission(inode2, MAY_WRITE))
1125 if (inode1->i_sb != inode2->i_sb)
1131 static int ll_swap_layouts_close(struct obd_client_handle *och,
1132 struct inode *inode, struct inode *inode2)
1134 const struct lu_fid *fid1 = ll_inode2fid(inode);
1135 const struct lu_fid *fid2;
1139 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1140 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1142 rc = ll_check_swap_layouts_validity(inode, inode2);
1144 GOTO(out_free_och, rc);
1146 /* We now know that inode2 is a lustre inode */
1147 fid2 = ll_inode2fid(inode2);
1149 rc = lu_fid_cmp(fid1, fid2);
1151 GOTO(out_free_och, rc = -EINVAL);
1153 /* Close the file and {swap,merge} layouts between inode & inode2.
1154 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1155 * because we still need it to pack l_remote_handle to MDT. */
1156 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1159 och = NULL; /* freed in ll_close_inode_openhandle() */
1169 * Release lease and close the file.
1170 * It will check if the lease has ever broken.
1172 static int ll_lease_close_intent(struct obd_client_handle *och,
1173 struct inode *inode,
1174 bool *lease_broken, enum mds_op_bias bias,
1177 struct ldlm_lock *lock;
1178 bool cancelled = true;
1182 lock = ldlm_handle2lock(&och->och_lease_handle);
1184 lock_res_and_lock(lock);
1185 cancelled = ldlm_is_cancel(lock);
1186 unlock_res_and_lock(lock);
1187 LDLM_LOCK_PUT(lock);
1190 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1191 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1193 if (lease_broken != NULL)
1194 *lease_broken = cancelled;
1196 if (!cancelled && !bias)
1197 ldlm_cli_cancel(&och->och_lease_handle, 0);
1199 if (cancelled) { /* no need to excute intent */
1204 rc = ll_close_inode_openhandle(inode, och, bias, data);
1208 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1211 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1215 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1217 static int ll_lease_file_resync(struct obd_client_handle *och,
1218 struct inode *inode, unsigned long arg)
1220 struct ll_sb_info *sbi = ll_i2sbi(inode);
1221 struct md_op_data *op_data;
1222 struct ll_ioc_lease_id ioc;
1223 __u64 data_version_unused;
1227 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1228 LUSTRE_OPC_ANY, NULL);
1229 if (IS_ERR(op_data))
1230 RETURN(PTR_ERR(op_data));
1232 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1236 /* before starting file resync, it's necessary to clean up page cache
1237 * in client memory, otherwise once the layout version is increased,
1238 * writing back cached data will be denied the OSTs. */
1239 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1243 op_data->op_lease_handle = och->och_lease_handle;
1244 op_data->op_mirror_id = ioc.lil_mirror_id;
1245 rc = md_file_resync(sbi->ll_md_exp, op_data);
1251 ll_finish_md_op_data(op_data);
1255 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1257 struct ll_inode_info *lli = ll_i2info(inode);
1258 struct cl_object *obj = lli->lli_clob;
1259 struct cl_attr *attr = vvp_env_thread_attr(env);
1267 ll_inode_size_lock(inode);
1269 /* Merge timestamps the most recently obtained from MDS with
1270 * timestamps obtained from OSTs.
1272 * Do not overwrite atime of inode because it may be refreshed
1273 * by file_accessed() function. If the read was served by cache
1274 * data, there is no RPC to be sent so that atime may not be
1275 * transferred to OSTs at all. MDT only updates atime at close time
1276 * if it's at least 'mdd.*.atime_diff' older.
1277 * All in all, the atime in Lustre does not strictly comply with
1278 * POSIX. Solving this problem needs to send an RPC to MDT for each
1279 * read, this will hurt performance.
1281 if (inode->i_atime.tv_sec < lli->lli_atime ||
1282 lli->lli_update_atime) {
1283 inode->i_atime.tv_sec = lli->lli_atime;
1284 lli->lli_update_atime = 0;
1286 inode->i_mtime.tv_sec = lli->lli_mtime;
1287 inode->i_ctime.tv_sec = lli->lli_ctime;
1289 mtime = inode->i_mtime.tv_sec;
1290 atime = inode->i_atime.tv_sec;
1291 ctime = inode->i_ctime.tv_sec;
1293 cl_object_attr_lock(obj);
1294 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1297 rc = cl_object_attr_get(env, obj, attr);
1298 cl_object_attr_unlock(obj);
1301 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1303 if (atime < attr->cat_atime)
1304 atime = attr->cat_atime;
1306 if (ctime < attr->cat_ctime)
1307 ctime = attr->cat_ctime;
1309 if (mtime < attr->cat_mtime)
1310 mtime = attr->cat_mtime;
1312 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1313 PFID(&lli->lli_fid), attr->cat_size);
1315 i_size_write(inode, attr->cat_size);
1316 inode->i_blocks = attr->cat_blocks;
1318 inode->i_mtime.tv_sec = mtime;
1319 inode->i_atime.tv_sec = atime;
1320 inode->i_ctime.tv_sec = ctime;
1323 ll_inode_size_unlock(inode);
1329 * Set designated mirror for I/O.
1331 * So far only read, write, and truncated can support to issue I/O to
1332 * designated mirror.
1334 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1336 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1338 /* clear layout version for generic(non-resync) I/O in case it carries
1339 * stale layout version due to I/O restart */
1340 io->ci_layout_version = 0;
1342 /* FLR: disable non-delay for designated mirror I/O because obviously
1343 * only one mirror is available */
1344 if (fd->fd_designated_mirror > 0) {
1346 io->ci_designated_mirror = fd->fd_designated_mirror;
1347 io->ci_layout_version = fd->fd_layout_version;
1350 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1351 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1354 static bool file_is_noatime(const struct file *file)
1356 const struct vfsmount *mnt = file->f_path.mnt;
1357 const struct inode *inode = file_inode((struct file *)file);
1359 /* Adapted from file_accessed() and touch_atime().*/
1360 if (file->f_flags & O_NOATIME)
1363 if (inode->i_flags & S_NOATIME)
1366 if (IS_NOATIME(inode))
1369 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1372 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1375 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1381 static void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot)
1383 struct inode *inode = file_inode(file);
1384 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1386 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1387 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1389 if (iot == CIT_WRITE) {
1390 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1391 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1392 file->f_flags & O_DIRECT ||
1395 io->ci_obj = ll_i2info(inode)->lli_clob;
1396 io->ci_lockreq = CILR_MAYBE;
1397 if (ll_file_nolock(file)) {
1398 io->ci_lockreq = CILR_NEVER;
1399 io->ci_no_srvlock = 1;
1400 } else if (file->f_flags & O_APPEND) {
1401 io->ci_lockreq = CILR_MANDATORY;
1403 io->ci_noatime = file_is_noatime(file);
1405 /* FLR: only use non-delay I/O for read as there is only one
1406 * avaliable mirror for write. */
1407 io->ci_ndelay = !(iot == CIT_WRITE);
1409 ll_io_set_mirror(io, file);
1412 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1415 struct ll_inode_info *lli = ll_i2info(inode);
1416 struct ll_sb_info *sbi = ll_i2sbi(inode);
1417 enum obd_heat_type sample_type;
1418 enum obd_heat_type iobyte_type;
1419 __u64 now = ktime_get_real_seconds();
1421 if (!ll_sbi_has_file_heat(sbi) ||
1422 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1425 if (iot == CIT_READ) {
1426 sample_type = OBD_HEAT_READSAMPLE;
1427 iobyte_type = OBD_HEAT_READBYTE;
1428 } else if (iot == CIT_WRITE) {
1429 sample_type = OBD_HEAT_WRITESAMPLE;
1430 iobyte_type = OBD_HEAT_WRITEBYTE;
1435 spin_lock(&lli->lli_heat_lock);
1436 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1437 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1438 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1439 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1440 spin_unlock(&lli->lli_heat_lock);
1444 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1445 struct file *file, enum cl_io_type iot,
1446 loff_t *ppos, size_t count)
1448 struct vvp_io *vio = vvp_env_io(env);
1449 struct inode *inode = file_inode(file);
1450 struct ll_inode_info *lli = ll_i2info(inode);
1451 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1452 struct range_lock range;
1456 unsigned retried = 0;
1457 bool restarted = false;
1461 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1462 file_dentry(file)->d_name.name,
1463 iot == CIT_READ ? "read" : "write", *ppos, count);
1466 io = vvp_env_thread_io(env);
1467 ll_io_init(io, file, iot);
1468 io->ci_ndelay_tried = retried;
1470 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1471 bool range_locked = false;
1473 if (file->f_flags & O_APPEND)
1474 range_lock_init(&range, 0, LUSTRE_EOF);
1476 range_lock_init(&range, *ppos, *ppos + count - 1);
1478 vio->vui_fd = LUSTRE_FPRIVATE(file);
1479 vio->vui_io_subtype = args->via_io_subtype;
1481 switch (vio->vui_io_subtype) {
1483 vio->vui_iter = args->u.normal.via_iter;
1484 vio->vui_iocb = args->u.normal.via_iocb;
1485 /* Direct IO reads must also take range lock,
1486 * or multiple reads will try to work on the same pages
1487 * See LU-6227 for details. */
1488 if (((iot == CIT_WRITE) ||
1489 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1490 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1491 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1493 rc = range_lock(&lli->lli_write_tree, &range);
1497 range_locked = true;
1501 vio->u.splice.vui_pipe = args->u.splice.via_pipe;
1502 vio->u.splice.vui_flags = args->u.splice.via_flags;
1505 CERROR("unknown IO subtype %u\n", vio->vui_io_subtype);
1509 ll_cl_add(file, env, io, LCC_RW);
1510 rc = cl_io_loop(env, io);
1511 ll_cl_remove(file, env);
1514 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1516 range_unlock(&lli->lli_write_tree, &range);
1519 /* cl_io_rw_init() handled IO */
1523 if (io->ci_nob > 0) {
1524 result += io->ci_nob;
1525 count -= io->ci_nob;
1526 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1528 /* prepare IO restart */
1529 if (count > 0 && args->via_io_subtype == IO_NORMAL)
1530 args->u.normal.via_iter = vio->vui_iter;
1533 cl_io_fini(env, io);
1536 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1537 file->f_path.dentry->d_name.name,
1538 iot, rc, result, io->ci_need_restart);
1540 if ((rc == 0 || rc == -ENODATA) && count > 0 && io->ci_need_restart) {
1542 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1543 file_dentry(file)->d_name.name,
1544 iot == CIT_READ ? "read" : "write",
1545 *ppos, count, result, rc);
1546 /* preserve the tried count for FLR */
1547 retried = io->ci_ndelay_tried;
1552 if (iot == CIT_READ) {
1554 ll_stats_ops_tally(ll_i2sbi(inode),
1555 LPROC_LL_READ_BYTES, result);
1556 } else if (iot == CIT_WRITE) {
1558 ll_stats_ops_tally(ll_i2sbi(inode),
1559 LPROC_LL_WRITE_BYTES, result);
1560 fd->fd_write_failed = false;
1561 } else if (result == 0 && rc == 0) {
1564 fd->fd_write_failed = true;
1566 fd->fd_write_failed = false;
1567 } else if (rc != -ERESTARTSYS) {
1568 fd->fd_write_failed = true;
1572 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1574 ll_heat_add(inode, iot, result);
1576 RETURN(result > 0 ? result : rc);
1580 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1581 * especially for small I/O.
1583 * To serve a read request, CLIO has to create and initialize a cl_io and
1584 * then request DLM lock. This has turned out to have siginificant overhead
1585 * and affects the performance of small I/O dramatically.
1587 * It's not necessary to create a cl_io for each I/O. Under the help of read
1588 * ahead, most of the pages being read are already in memory cache and we can
1589 * read those pages directly because if the pages exist, the corresponding DLM
1590 * lock must exist so that page content must be valid.
1592 * In fast read implementation, the llite speculatively finds and reads pages
1593 * in memory cache. There are three scenarios for fast read:
1594 * - If the page exists and is uptodate, kernel VM will provide the data and
1595 * CLIO won't be intervened;
1596 * - If the page was brought into memory by read ahead, it will be exported
1597 * and read ahead parameters will be updated;
1598 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1599 * it will go back and invoke normal read, i.e., a cl_io will be created
1600 * and DLM lock will be requested.
1602 * POSIX compliance: posix standard states that read is intended to be atomic.
1603 * Lustre read implementation is in line with Linux kernel read implementation
1604 * and neither of them complies with POSIX standard in this matter. Fast read
1605 * doesn't make the situation worse on single node but it may interleave write
1606 * results from multiple nodes due to short read handling in ll_file_aio_read().
1608 * \param env - lu_env
1609 * \param iocb - kiocb from kernel
1610 * \param iter - user space buffers where the data will be copied
1612 * \retval - number of bytes have been read, or error code if error occurred.
1615 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1619 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1622 /* NB: we can't do direct IO for fast read because it will need a lock
1623 * to make IO engine happy. */
1624 if (iocb->ki_filp->f_flags & O_DIRECT)
1627 result = generic_file_read_iter(iocb, iter);
1629 /* If the first page is not in cache, generic_file_aio_read() will be
1630 * returned with -ENODATA.
1631 * See corresponding code in ll_readpage(). */
1632 if (result == -ENODATA)
1636 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1637 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1638 LPROC_LL_READ_BYTES, result);
1645 * Read from a file (through the page cache).
1647 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1650 struct vvp_io_args *args;
1657 * Currently when PCC read failed, we do not fall back to the
1658 * normal read path, just return the error.
1659 * The resaon is that: for RW-PCC, the file data may be modified
1660 * in the PCC and inconsistent with the data on OSTs (or file
1661 * data has been removed from the Lustre file system), at this
1662 * time, fallback to the normal read path may read the wrong
1664 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1665 * path: read data from data copy on OSTs.
1667 result = pcc_file_read_iter(iocb, to, &cached);
1671 ll_ras_enter(iocb->ki_filp);
1673 result = ll_do_fast_read(iocb, to);
1674 if (result < 0 || iov_iter_count(to) == 0)
1677 env = cl_env_get(&refcheck);
1679 return PTR_ERR(env);
1681 args = ll_env_args(env, IO_NORMAL);
1682 args->u.normal.via_iter = to;
1683 args->u.normal.via_iocb = iocb;
1685 rc2 = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1686 &iocb->ki_pos, iov_iter_count(to));
1689 else if (result == 0)
1692 cl_env_put(env, &refcheck);
1698 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
1699 * If a page is already in the page cache and dirty (and some other things -
1700 * See ll_tiny_write_begin for the instantiation of these rules), then we can
1701 * write to it without doing a full I/O, because Lustre already knows about it
1702 * and will write it out. This saves a lot of processing time.
1704 * All writes here are within one page, so exclusion is handled by the page
1705 * lock on the vm page. We do not do tiny writes for writes which touch
1706 * multiple pages because it's very unlikely multiple sequential pages are
1707 * are already dirty.
1709 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
1710 * and are unlikely to be to already dirty pages.
1712 * Attribute updates are important here, we do them in ll_tiny_write_end.
1714 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
1716 ssize_t count = iov_iter_count(iter);
1717 struct file *file = iocb->ki_filp;
1718 struct inode *inode = file_inode(file);
1719 bool lock_inode = !IS_NOSEC(inode);
1724 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
1725 * of function for why.
1727 if (count >= PAGE_SIZE ||
1728 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
1731 if (unlikely(lock_inode))
1733 result = __generic_file_write_iter(iocb, iter);
1735 if (unlikely(lock_inode))
1736 inode_unlock(inode);
1738 /* If the page is not already dirty, ll_tiny_write_begin returns
1739 * -ENODATA. We continue on to normal write.
1741 if (result == -ENODATA)
1745 ll_heat_add(inode, CIT_WRITE, result);
1746 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES,
1748 ll_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED);
1751 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
1757 * Write to a file (through the page cache).
1759 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1761 struct vvp_io_args *args;
1763 ssize_t rc_tiny = 0, rc_normal;
1771 * When PCC write failed, we usually do not fall back to the normal
1772 * write path, just return the error. But there is a special case when
1773 * returned error code is -ENOSPC due to running out of space on PCC HSM
1774 * bakcend. At this time, it will fall back to normal I/O path and
1775 * retry the I/O. As the file is in HSM released state, it will restore
1776 * the file data to OSTs first and redo the write again. And the
1777 * restore process will revoke the layout lock and detach the file
1778 * from PCC cache automatically.
1780 result = pcc_file_write_iter(iocb, from, &cached);
1781 if (cached && result != -ENOSPC && result != -EDQUOT)
1784 /* NB: we can't do direct IO for tiny writes because they use the page
1785 * cache, we can't do sync writes because tiny writes can't flush
1786 * pages, and we can't do append writes because we can't guarantee the
1787 * required DLM locks are held to protect file size.
1789 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(iocb->ki_filp))) &&
1790 !(iocb->ki_filp->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
1791 rc_tiny = ll_do_tiny_write(iocb, from);
1793 /* In case of error, go on and try normal write - Only stop if tiny
1794 * write completed I/O.
1796 if (iov_iter_count(from) == 0)
1797 GOTO(out, rc_normal = rc_tiny);
1799 env = cl_env_get(&refcheck);
1801 return PTR_ERR(env);
1803 args = ll_env_args(env, IO_NORMAL);
1804 args->u.normal.via_iter = from;
1805 args->u.normal.via_iocb = iocb;
1807 rc_normal = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1808 &iocb->ki_pos, iov_iter_count(from));
1810 /* On success, combine bytes written. */
1811 if (rc_tiny >= 0 && rc_normal > 0)
1812 rc_normal += rc_tiny;
1813 /* On error, only return error from normal write if tiny write did not
1814 * write any bytes. Otherwise return bytes written by tiny write.
1816 else if (rc_tiny > 0)
1817 rc_normal = rc_tiny;
1819 cl_env_put(env, &refcheck);
1824 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1826 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1828 static int ll_file_get_iov_count(const struct iovec *iov,
1829 unsigned long *nr_segs, size_t *count)
1834 for (seg = 0; seg < *nr_segs; seg++) {
1835 const struct iovec *iv = &iov[seg];
1838 * If any segment has a negative length, or the cumulative
1839 * length ever wraps negative then return -EINVAL.
1842 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1844 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
1849 cnt -= iv->iov_len; /* This segment is no good */
1856 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1857 unsigned long nr_segs, loff_t pos)
1864 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count);
1868 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1869 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
1870 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1871 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
1872 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1874 result = ll_file_read_iter(iocb, &to);
1879 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
1882 struct iovec iov = { .iov_base = buf, .iov_len = count };
1887 init_sync_kiocb(&kiocb, file);
1888 kiocb.ki_pos = *ppos;
1889 #ifdef HAVE_KIOCB_KI_LEFT
1890 kiocb.ki_left = count;
1891 #elif defined(HAVE_KI_NBYTES)
1892 kiocb.i_nbytes = count;
1895 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
1896 *ppos = kiocb.ki_pos;
1902 * Write to a file (through the page cache).
1905 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1906 unsigned long nr_segs, loff_t pos)
1908 struct iov_iter from;
1913 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count);
1917 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1918 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
1919 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1920 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
1921 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1923 result = ll_file_write_iter(iocb, &from);
1928 static ssize_t ll_file_write(struct file *file, const char __user *buf,
1929 size_t count, loff_t *ppos)
1931 struct iovec iov = { .iov_base = (void __user *)buf,
1938 init_sync_kiocb(&kiocb, file);
1939 kiocb.ki_pos = *ppos;
1940 #ifdef HAVE_KIOCB_KI_LEFT
1941 kiocb.ki_left = count;
1942 #elif defined(HAVE_KI_NBYTES)
1943 kiocb.ki_nbytes = count;
1946 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
1947 *ppos = kiocb.ki_pos;
1951 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
1954 * Send file content (through pagecache) somewhere with helper
1956 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1957 struct pipe_inode_info *pipe, size_t count,
1961 struct vvp_io_args *args;
1968 result = pcc_file_splice_read(in_file, ppos, pipe,
1969 count, flags, &cached);
1973 ll_ras_enter(in_file);
1975 env = cl_env_get(&refcheck);
1977 RETURN(PTR_ERR(env));
1979 args = ll_env_args(env, IO_SPLICE);
1980 args->u.splice.via_pipe = pipe;
1981 args->u.splice.via_flags = flags;
1983 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1984 cl_env_put(env, &refcheck);
1988 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1989 __u64 flags, struct lov_user_md *lum, int lum_size)
1991 struct lookup_intent oit = {
1993 .it_flags = flags | MDS_OPEN_BY_FID,
1998 ll_inode_size_lock(inode);
1999 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2001 GOTO(out_unlock, rc);
2003 ll_release_openhandle(dentry, &oit);
2006 ll_inode_size_unlock(inode);
2007 ll_intent_release(&oit);
2012 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2013 struct lov_mds_md **lmmp, int *lmm_size,
2014 struct ptlrpc_request **request)
2016 struct ll_sb_info *sbi = ll_i2sbi(inode);
2017 struct mdt_body *body;
2018 struct lov_mds_md *lmm = NULL;
2019 struct ptlrpc_request *req = NULL;
2020 struct md_op_data *op_data;
2023 rc = ll_get_default_mdsize(sbi, &lmmsize);
2027 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2028 strlen(filename), lmmsize,
2029 LUSTRE_OPC_ANY, NULL);
2030 if (IS_ERR(op_data))
2031 RETURN(PTR_ERR(op_data));
2033 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2034 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2035 ll_finish_md_op_data(op_data);
2037 CDEBUG(D_INFO, "md_getattr_name failed "
2038 "on %s: rc %d\n", filename, rc);
2042 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2043 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2045 lmmsize = body->mbo_eadatasize;
2047 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2049 GOTO(out, rc = -ENODATA);
2052 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2053 LASSERT(lmm != NULL);
2055 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2056 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2057 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2058 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2059 GOTO(out, rc = -EPROTO);
2062 * This is coming from the MDS, so is probably in
2063 * little endian. We convert it to host endian before
2064 * passing it to userspace.
2066 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
2069 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2070 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2071 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2072 if (le32_to_cpu(lmm->lmm_pattern) &
2073 LOV_PATTERN_F_RELEASED)
2077 /* if function called for directory - we should
2078 * avoid swab not existent lsm objects */
2079 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
2080 lustre_swab_lov_user_md_v1(
2081 (struct lov_user_md_v1 *)lmm);
2082 if (S_ISREG(body->mbo_mode))
2083 lustre_swab_lov_user_md_objects(
2084 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2086 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2087 lustre_swab_lov_user_md_v3(
2088 (struct lov_user_md_v3 *)lmm);
2089 if (S_ISREG(body->mbo_mode))
2090 lustre_swab_lov_user_md_objects(
2091 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2093 } else if (lmm->lmm_magic ==
2094 cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2095 lustre_swab_lov_comp_md_v1(
2096 (struct lov_comp_md_v1 *)lmm);
2097 } else if (lmm->lmm_magic ==
2098 cpu_to_le32(LOV_MAGIC_FOREIGN)) {
2099 struct lov_foreign_md *lfm;
2101 lfm = (struct lov_foreign_md *)lmm;
2102 __swab32s(&lfm->lfm_magic);
2103 __swab32s(&lfm->lfm_length);
2104 __swab32s(&lfm->lfm_type);
2105 __swab32s(&lfm->lfm_flags);
2111 *lmm_size = lmmsize;
2116 static int ll_lov_setea(struct inode *inode, struct file *file,
2119 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2120 struct lov_user_md *lump;
2121 int lum_size = sizeof(struct lov_user_md) +
2122 sizeof(struct lov_user_ost_data);
2126 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2129 OBD_ALLOC_LARGE(lump, lum_size);
2133 if (copy_from_user(lump, arg, lum_size))
2134 GOTO(out_lump, rc = -EFAULT);
2136 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2138 cl_lov_delay_create_clear(&file->f_flags);
2141 OBD_FREE_LARGE(lump, lum_size);
2145 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2152 env = cl_env_get(&refcheck);
2154 RETURN(PTR_ERR(env));
2156 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2157 cl_env_put(env, &refcheck);
2161 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2164 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2165 struct lov_user_md *klum;
2167 __u64 flags = FMODE_WRITE;
2170 rc = ll_copy_user_md(lum, &klum);
2175 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2180 rc = put_user(0, &lum->lmm_stripe_count);
2184 rc = ll_layout_refresh(inode, &gen);
2188 rc = ll_file_getstripe(inode, arg, lum_size);
2190 cl_lov_delay_create_clear(&file->f_flags);
2193 OBD_FREE(klum, lum_size);
2198 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2200 struct ll_inode_info *lli = ll_i2info(inode);
2201 struct cl_object *obj = lli->lli_clob;
2202 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2203 struct ll_grouplock grouplock;
2208 CWARN("group id for group lock must not be 0\n");
2212 if (ll_file_nolock(file))
2213 RETURN(-EOPNOTSUPP);
2215 spin_lock(&lli->lli_lock);
2216 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2217 CWARN("group lock already existed with gid %lu\n",
2218 fd->fd_grouplock.lg_gid);
2219 spin_unlock(&lli->lli_lock);
2222 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2223 spin_unlock(&lli->lli_lock);
2226 * XXX: group lock needs to protect all OST objects while PFL
2227 * can add new OST objects during the IO, so we'd instantiate
2228 * all OST objects before getting its group lock.
2233 struct cl_layout cl = {
2234 .cl_is_composite = false,
2236 struct lu_extent ext = {
2238 .e_end = OBD_OBJECT_EOF,
2241 env = cl_env_get(&refcheck);
2243 RETURN(PTR_ERR(env));
2245 rc = cl_object_layout_get(env, obj, &cl);
2246 if (!rc && cl.cl_is_composite)
2247 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2250 cl_env_put(env, &refcheck);
2255 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2256 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2260 spin_lock(&lli->lli_lock);
2261 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2262 spin_unlock(&lli->lli_lock);
2263 CERROR("another thread just won the race\n");
2264 cl_put_grouplock(&grouplock);
2268 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2269 fd->fd_grouplock = grouplock;
2270 spin_unlock(&lli->lli_lock);
2272 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2276 static int ll_put_grouplock(struct inode *inode, struct file *file,
2279 struct ll_inode_info *lli = ll_i2info(inode);
2280 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2281 struct ll_grouplock grouplock;
2284 spin_lock(&lli->lli_lock);
2285 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2286 spin_unlock(&lli->lli_lock);
2287 CWARN("no group lock held\n");
2291 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2293 if (fd->fd_grouplock.lg_gid != arg) {
2294 CWARN("group lock %lu doesn't match current id %lu\n",
2295 arg, fd->fd_grouplock.lg_gid);
2296 spin_unlock(&lli->lli_lock);
2300 grouplock = fd->fd_grouplock;
2301 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2302 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2303 spin_unlock(&lli->lli_lock);
2305 cl_put_grouplock(&grouplock);
2306 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2311 * Close inode open handle
2313 * \param dentry [in] dentry which contains the inode
2314 * \param it [in,out] intent which contains open info and result
2317 * \retval <0 failure
2319 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2321 struct inode *inode = dentry->d_inode;
2322 struct obd_client_handle *och;
2328 /* Root ? Do nothing. */
2329 if (dentry->d_inode->i_sb->s_root == dentry)
2332 /* No open handle to close? Move away */
2333 if (!it_disposition(it, DISP_OPEN_OPEN))
2336 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2338 OBD_ALLOC(och, sizeof(*och));
2340 GOTO(out, rc = -ENOMEM);
2342 ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2344 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2346 /* this one is in place of ll_file_open */
2347 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2348 ptlrpc_req_finished(it->it_request);
2349 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2355 * Get size for inode for which FIEMAP mapping is requested.
2356 * Make the FIEMAP get_info call and returns the result.
2357 * \param fiemap kernel buffer to hold extens
2358 * \param num_bytes kernel buffer size
2360 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2366 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2369 /* Checks for fiemap flags */
2370 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2371 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2375 /* Check for FIEMAP_FLAG_SYNC */
2376 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2377 rc = filemap_fdatawrite(inode->i_mapping);
2382 env = cl_env_get(&refcheck);
2384 RETURN(PTR_ERR(env));
2386 if (i_size_read(inode) == 0) {
2387 rc = ll_glimpse_size(inode);
2392 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2393 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2394 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2396 /* If filesize is 0, then there would be no objects for mapping */
2397 if (fmkey.lfik_oa.o_size == 0) {
2398 fiemap->fm_mapped_extents = 0;
2402 fmkey.lfik_fiemap = *fiemap;
2404 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2405 &fmkey, fiemap, &num_bytes);
2407 cl_env_put(env, &refcheck);
2411 int ll_fid2path(struct inode *inode, void __user *arg)
2413 struct obd_export *exp = ll_i2mdexp(inode);
2414 const struct getinfo_fid2path __user *gfin = arg;
2416 struct getinfo_fid2path *gfout;
2422 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2423 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2426 /* Only need to get the buflen */
2427 if (get_user(pathlen, &gfin->gf_pathlen))
2430 if (pathlen > PATH_MAX)
2433 outsize = sizeof(*gfout) + pathlen;
2434 OBD_ALLOC(gfout, outsize);
2438 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2439 GOTO(gf_free, rc = -EFAULT);
2440 /* append root FID after gfout to let MDT know the root FID so that it
2441 * can lookup the correct path, this is mainly for fileset.
2442 * old server without fileset mount support will ignore this. */
2443 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2445 /* Call mdc_iocontrol */
2446 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2450 if (copy_to_user(arg, gfout, outsize))
2454 OBD_FREE(gfout, outsize);
2459 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2461 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2469 ioc->idv_version = 0;
2470 ioc->idv_layout_version = UINT_MAX;
2472 /* If no file object initialized, we consider its version is 0. */
2476 env = cl_env_get(&refcheck);
2478 RETURN(PTR_ERR(env));
2480 io = vvp_env_thread_io(env);
2482 io->u.ci_data_version.dv_data_version = 0;
2483 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2484 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2487 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2488 result = cl_io_loop(env, io);
2490 result = io->ci_result;
2492 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2493 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2495 cl_io_fini(env, io);
2497 if (unlikely(io->ci_need_restart))
2500 cl_env_put(env, &refcheck);
2506 * Read the data_version for inode.
2508 * This value is computed using stripe object version on OST.
2509 * Version is computed using server side locking.
2511 * @param flags if do sync on the OST side;
2513 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2514 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2516 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2518 struct ioc_data_version ioc = { .idv_flags = flags };
2521 rc = ll_ioc_data_version(inode, &ioc);
2523 *data_version = ioc.idv_version;
2529 * Trigger a HSM release request for the provided inode.
2531 int ll_hsm_release(struct inode *inode)
2534 struct obd_client_handle *och = NULL;
2535 __u64 data_version = 0;
2540 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2541 ll_i2sbi(inode)->ll_fsname,
2542 PFID(&ll_i2info(inode)->lli_fid));
2544 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2546 GOTO(out, rc = PTR_ERR(och));
2548 /* Grab latest data_version and [am]time values */
2549 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2553 env = cl_env_get(&refcheck);
2555 GOTO(out, rc = PTR_ERR(env));
2557 rc = ll_merge_attr(env, inode);
2558 cl_env_put(env, &refcheck);
2560 /* If error happen, we have the wrong size for a file.
2566 /* Release the file.
2567 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2568 * we still need it to pack l_remote_handle to MDT. */
2569 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2575 if (och != NULL && !IS_ERR(och)) /* close the file */
2576 ll_lease_close(och, inode, NULL);
2581 struct ll_swap_stack {
2584 struct inode *inode1;
2585 struct inode *inode2;
2590 static int ll_swap_layouts(struct file *file1, struct file *file2,
2591 struct lustre_swap_layouts *lsl)
2593 struct mdc_swap_layouts msl;
2594 struct md_op_data *op_data;
2597 struct ll_swap_stack *llss = NULL;
2600 OBD_ALLOC_PTR(llss);
2604 llss->inode1 = file_inode(file1);
2605 llss->inode2 = file_inode(file2);
2607 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2611 /* we use 2 bool because it is easier to swap than 2 bits */
2612 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2613 llss->check_dv1 = true;
2615 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2616 llss->check_dv2 = true;
2618 /* we cannot use lsl->sl_dvX directly because we may swap them */
2619 llss->dv1 = lsl->sl_dv1;
2620 llss->dv2 = lsl->sl_dv2;
2622 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2623 if (rc == 0) /* same file, done! */
2626 if (rc < 0) { /* sequentialize it */
2627 swap(llss->inode1, llss->inode2);
2629 swap(llss->dv1, llss->dv2);
2630 swap(llss->check_dv1, llss->check_dv2);
2634 if (gid != 0) { /* application asks to flush dirty cache */
2635 rc = ll_get_grouplock(llss->inode1, file1, gid);
2639 rc = ll_get_grouplock(llss->inode2, file2, gid);
2641 ll_put_grouplock(llss->inode1, file1, gid);
2646 /* ultimate check, before swaping the layouts we check if
2647 * dataversion has changed (if requested) */
2648 if (llss->check_dv1) {
2649 rc = ll_data_version(llss->inode1, &dv, 0);
2652 if (dv != llss->dv1)
2653 GOTO(putgl, rc = -EAGAIN);
2656 if (llss->check_dv2) {
2657 rc = ll_data_version(llss->inode2, &dv, 0);
2660 if (dv != llss->dv2)
2661 GOTO(putgl, rc = -EAGAIN);
2664 /* struct md_op_data is used to send the swap args to the mdt
2665 * only flags is missing, so we use struct mdc_swap_layouts
2666 * through the md_op_data->op_data */
2667 /* flags from user space have to be converted before they are send to
2668 * server, no flag is sent today, they are only used on the client */
2671 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2672 0, LUSTRE_OPC_ANY, &msl);
2673 if (IS_ERR(op_data))
2674 GOTO(free, rc = PTR_ERR(op_data));
2676 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2677 sizeof(*op_data), op_data, NULL);
2678 ll_finish_md_op_data(op_data);
2685 ll_put_grouplock(llss->inode2, file2, gid);
2686 ll_put_grouplock(llss->inode1, file1, gid);
2696 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2698 struct obd_export *exp = ll_i2mdexp(inode);
2699 struct md_op_data *op_data;
2703 /* Detect out-of range masks */
2704 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2707 /* Non-root users are forbidden to set or clear flags which are
2708 * NOT defined in HSM_USER_MASK. */
2709 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2710 !cfs_capable(CFS_CAP_SYS_ADMIN))
2713 if (!exp_connect_archive_id_array(exp)) {
2714 /* Detect out-of range archive id */
2715 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2716 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
2720 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2721 LUSTRE_OPC_ANY, hss);
2722 if (IS_ERR(op_data))
2723 RETURN(PTR_ERR(op_data));
2725 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
2728 ll_finish_md_op_data(op_data);
2733 static int ll_hsm_import(struct inode *inode, struct file *file,
2734 struct hsm_user_import *hui)
2736 struct hsm_state_set *hss = NULL;
2737 struct iattr *attr = NULL;
2741 if (!S_ISREG(inode->i_mode))
2747 GOTO(out, rc = -ENOMEM);
2749 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2750 hss->hss_archive_id = hui->hui_archive_id;
2751 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2752 rc = ll_hsm_state_set(inode, hss);
2756 OBD_ALLOC_PTR(attr);
2758 GOTO(out, rc = -ENOMEM);
2760 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2761 attr->ia_mode |= S_IFREG;
2762 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2763 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2764 attr->ia_size = hui->hui_size;
2765 attr->ia_mtime.tv_sec = hui->hui_mtime;
2766 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2767 attr->ia_atime.tv_sec = hui->hui_atime;
2768 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2770 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2771 ATTR_UID | ATTR_GID |
2772 ATTR_MTIME | ATTR_MTIME_SET |
2773 ATTR_ATIME | ATTR_ATIME_SET;
2777 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
2781 inode_unlock(inode);
2793 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2795 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2796 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2799 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
2801 struct inode *inode = file_inode(file);
2803 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
2804 ATTR_MTIME | ATTR_MTIME_SET |
2807 .tv_sec = lfu->lfu_atime_sec,
2808 .tv_nsec = lfu->lfu_atime_nsec,
2811 .tv_sec = lfu->lfu_mtime_sec,
2812 .tv_nsec = lfu->lfu_mtime_nsec,
2815 .tv_sec = lfu->lfu_ctime_sec,
2816 .tv_nsec = lfu->lfu_ctime_nsec,
2822 if (!capable(CAP_SYS_ADMIN))
2825 if (!S_ISREG(inode->i_mode))
2829 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
2831 inode_unlock(inode);
2836 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
2839 case MODE_READ_USER:
2841 case MODE_WRITE_USER:
2848 static const char *const user_lockname[] = LOCK_MODE_NAMES;
2850 /* Used to allow the upper layers of the client to request an LDLM lock
2851 * without doing an actual read or write.
2853 * Used for ladvise lockahead to manually request specific locks.
2855 * \param[in] file file this ladvise lock request is on
2856 * \param[in] ladvise ladvise struct describing this lock request
2858 * \retval 0 success, no detailed result available (sync requests
2859 * and requests sent to the server [not handled locally]
2860 * cannot return detailed results)
2861 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
2862 * see definitions for details.
2863 * \retval negative negative errno on error
2865 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
2867 struct lu_env *env = NULL;
2868 struct cl_io *io = NULL;
2869 struct cl_lock *lock = NULL;
2870 struct cl_lock_descr *descr = NULL;
2871 struct dentry *dentry = file->f_path.dentry;
2872 struct inode *inode = dentry->d_inode;
2873 enum cl_lock_mode cl_mode;
2874 off_t start = ladvise->lla_start;
2875 off_t end = ladvise->lla_end;
2881 CDEBUG(D_VFSTRACE, "Lock request: file=%.*s, inode=%p, mode=%s "
2882 "start=%llu, end=%llu\n", dentry->d_name.len,
2883 dentry->d_name.name, dentry->d_inode,
2884 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
2887 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
2889 GOTO(out, result = cl_mode);
2891 /* Get IO environment */
2892 result = cl_io_get(inode, &env, &io, &refcheck);
2896 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
2899 * nothing to do for this io. This currently happens when
2900 * stripe sub-object's are not yet created.
2902 result = io->ci_result;
2903 } else if (result == 0) {
2904 lock = vvp_env_lock(env);
2905 descr = &lock->cll_descr;
2907 descr->cld_obj = io->ci_obj;
2908 /* Convert byte offsets to pages */
2909 descr->cld_start = cl_index(io->ci_obj, start);
2910 descr->cld_end = cl_index(io->ci_obj, end);
2911 descr->cld_mode = cl_mode;
2912 /* CEF_MUST is used because we do not want to convert a
2913 * lockahead request to a lockless lock */
2914 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND |
2917 if (ladvise->lla_peradvice_flags & LF_ASYNC)
2918 descr->cld_enq_flags |= CEF_SPECULATIVE;
2920 result = cl_lock_request(env, io, lock);
2922 /* On success, we need to release the lock */
2924 cl_lock_release(env, lock);
2926 cl_io_fini(env, io);
2927 cl_env_put(env, &refcheck);
2929 /* -ECANCELED indicates a matching lock with a different extent
2930 * was already present, and -EEXIST indicates a matching lock
2931 * on exactly the same extent was already present.
2932 * We convert them to positive values for userspace to make
2933 * recognizing true errors easier.
2934 * Note we can only return these detailed results on async requests,
2935 * as sync requests look the same as i/o requests for locking. */
2936 if (result == -ECANCELED)
2937 result = LLA_RESULT_DIFFERENT;
2938 else if (result == -EEXIST)
2939 result = LLA_RESULT_SAME;
2944 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
2946 static int ll_ladvise_sanity(struct inode *inode,
2947 struct llapi_lu_ladvise *ladvise)
2949 struct ll_sb_info *sbi = ll_i2sbi(inode);
2950 enum lu_ladvise_type advice = ladvise->lla_advice;
2951 /* Note the peradvice flags is a 32 bit field, so per advice flags must
2952 * be in the first 32 bits of enum ladvise_flags */
2953 __u32 flags = ladvise->lla_peradvice_flags;
2954 /* 3 lines at 80 characters per line, should be plenty */
2957 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
2959 CDEBUG(D_VFSTRACE, "%s: advice with value '%d' not recognized,"
2960 "last supported advice is %s (value '%d'): rc = %d\n",
2961 sbi->ll_fsname, advice,
2962 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
2966 /* Per-advice checks */
2968 case LU_LADVISE_LOCKNOEXPAND:
2969 if (flags & ~LF_LOCKNOEXPAND_MASK) {
2971 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
2972 "rc = %d\n", sbi->ll_fsname, flags,
2973 ladvise_names[advice], rc);
2977 case LU_LADVISE_LOCKAHEAD:
2978 /* Currently only READ and WRITE modes can be requested */
2979 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
2980 ladvise->lla_lockahead_mode == 0) {
2982 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
2983 "rc = %d\n", sbi->ll_fsname,
2984 ladvise->lla_lockahead_mode,
2985 ladvise_names[advice], rc);
2988 case LU_LADVISE_WILLREAD:
2989 case LU_LADVISE_DONTNEED:
2991 /* Note fall through above - These checks apply to all advices
2992 * except LOCKNOEXPAND */
2993 if (flags & ~LF_DEFAULT_MASK) {
2995 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
2996 "rc = %d\n", sbi->ll_fsname, flags,
2997 ladvise_names[advice], rc);
3000 if (ladvise->lla_start >= ladvise->lla_end) {
3002 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3003 "for %s: rc = %d\n", sbi->ll_fsname,
3004 ladvise->lla_start, ladvise->lla_end,
3005 ladvise_names[advice], rc);
3017 * Give file access advices
3019 * The ladvise interface is similar to Linux fadvise() system call, except it
3020 * forwards the advices directly from Lustre client to server. The server side
3021 * codes will apply appropriate read-ahead and caching techniques for the
3022 * corresponding files.
3024 * A typical workload for ladvise is e.g. a bunch of different clients are
3025 * doing small random reads of a file, so prefetching pages into OSS cache
3026 * with big linear reads before the random IO is a net benefit. Fetching
3027 * all that data into each client cache with fadvise() may not be, due to
3028 * much more data being sent to the client.
3030 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3031 struct llapi_lu_ladvise *ladvise)
3035 struct cl_ladvise_io *lio;
3040 env = cl_env_get(&refcheck);
3042 RETURN(PTR_ERR(env));
3044 io = vvp_env_thread_io(env);
3045 io->ci_obj = ll_i2info(inode)->lli_clob;
3047 /* initialize parameters for ladvise */
3048 lio = &io->u.ci_ladvise;
3049 lio->li_start = ladvise->lla_start;
3050 lio->li_end = ladvise->lla_end;
3051 lio->li_fid = ll_inode2fid(inode);
3052 lio->li_advice = ladvise->lla_advice;
3053 lio->li_flags = flags;
3055 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3056 rc = cl_io_loop(env, io);
3060 cl_io_fini(env, io);
3061 cl_env_put(env, &refcheck);
3065 static int ll_lock_noexpand(struct file *file, int flags)
3067 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3069 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3074 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3077 struct fsxattr fsxattr;
3079 if (copy_from_user(&fsxattr,
3080 (const struct fsxattr __user *)arg,
3084 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3085 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
3086 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3087 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3088 if (copy_to_user((struct fsxattr __user *)arg,
3089 &fsxattr, sizeof(fsxattr)))
3095 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa)
3098 * Project Quota ID state is only allowed to change from within the init
3099 * namespace. Enforce that restriction only if we are trying to change
3100 * the quota ID state. Everything else is allowed in user namespaces.
3102 if (current_user_ns() == &init_user_ns)
3105 if (ll_i2info(inode)->lli_projid != fa->fsx_projid)
3108 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT)) {
3109 if (!(fa->fsx_xflags & FS_XFLAG_PROJINHERIT))
3112 if (fa->fsx_xflags & FS_XFLAG_PROJINHERIT)
3119 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3123 struct md_op_data *op_data;
3124 struct ptlrpc_request *req = NULL;
3126 struct fsxattr fsxattr;
3127 struct cl_object *obj;
3131 if (copy_from_user(&fsxattr,
3132 (const struct fsxattr __user *)arg,
3136 rc = ll_ioctl_check_project(inode, &fsxattr);
3140 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3141 LUSTRE_OPC_ANY, NULL);
3142 if (IS_ERR(op_data))
3143 RETURN(PTR_ERR(op_data));
3145 flags = ll_xflags_to_inode_flags(fsxattr.fsx_xflags);
3146 op_data->op_attr_flags = ll_inode_to_ext_flags(flags);
3147 if (fsxattr.fsx_xflags & FS_XFLAG_PROJINHERIT)
3148 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3149 op_data->op_projid = fsxattr.fsx_projid;
3150 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3151 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL,
3153 ptlrpc_req_finished(req);
3155 GOTO(out_fsxattr, rc);
3156 ll_update_inode_flags(inode, op_data->op_attr_flags);
3157 obj = ll_i2info(inode)->lli_clob;
3159 GOTO(out_fsxattr, rc);
3161 OBD_ALLOC_PTR(attr);
3163 GOTO(out_fsxattr, rc = -ENOMEM);
3165 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS,
3166 fsxattr.fsx_xflags);
3169 ll_finish_md_op_data(op_data);
3173 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3176 struct inode *inode = file_inode(file);
3177 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3178 struct ll_inode_info *lli = ll_i2info(inode);
3179 struct obd_client_handle *och = NULL;
3180 struct split_param sp;
3181 struct pcc_param param;
3182 bool lease_broken = false;
3184 enum mds_op_bias bias = 0;
3185 struct file *layout_file = NULL;
3187 size_t data_size = 0;
3188 bool attached = false;
3193 mutex_lock(&lli->lli_och_mutex);
3194 if (fd->fd_lease_och != NULL) {
3195 och = fd->fd_lease_och;
3196 fd->fd_lease_och = NULL;
3198 mutex_unlock(&lli->lli_och_mutex);
3203 fmode = och->och_flags;
3205 switch (ioc->lil_flags) {
3206 case LL_LEASE_RESYNC_DONE:
3207 if (ioc->lil_count > IOC_IDS_MAX)
3208 GOTO(out_lease_close, rc = -EINVAL);
3210 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3211 OBD_ALLOC(data, data_size);
3213 GOTO(out_lease_close, rc = -ENOMEM);
3215 if (copy_from_user(data, (void __user *)arg, data_size))
3216 GOTO(out_lease_close, rc = -EFAULT);
3218 bias = MDS_CLOSE_RESYNC_DONE;
3220 case LL_LEASE_LAYOUT_MERGE: {
3223 if (ioc->lil_count != 1)
3224 GOTO(out_lease_close, rc = -EINVAL);
3226 arg += sizeof(*ioc);
3227 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3228 GOTO(out_lease_close, rc = -EFAULT);
3230 layout_file = fget(fd);
3232 GOTO(out_lease_close, rc = -EBADF);
3234 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3235 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3236 GOTO(out_lease_close, rc = -EPERM);
3238 data = file_inode(layout_file);
3239 bias = MDS_CLOSE_LAYOUT_MERGE;
3242 case LL_LEASE_LAYOUT_SPLIT: {
3246 if (ioc->lil_count != 2)
3247 GOTO(out_lease_close, rc = -EINVAL);
3249 arg += sizeof(*ioc);
3250 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3251 GOTO(out_lease_close, rc = -EFAULT);
3253 arg += sizeof(__u32);
3254 if (copy_from_user(&mirror_id, (void __user *)arg,
3256 GOTO(out_lease_close, rc = -EFAULT);
3258 layout_file = fget(fdv);
3260 GOTO(out_lease_close, rc = -EBADF);
3262 sp.sp_inode = file_inode(layout_file);
3263 sp.sp_mirror_id = (__u16)mirror_id;
3265 bias = MDS_CLOSE_LAYOUT_SPLIT;
3268 case LL_LEASE_PCC_ATTACH:
3269 if (ioc->lil_count != 1)
3272 arg += sizeof(*ioc);
3273 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3275 GOTO(out_lease_close, rc2 = -EFAULT);
3277 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3279 GOTO(out_lease_close, rc2);
3282 /* Grab latest data version */
3283 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3286 GOTO(out_lease_close, rc2);
3289 bias = MDS_PCC_ATTACH;
3292 /* without close intent */
3297 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3301 rc = ll_lease_och_release(inode, file);
3310 switch (ioc->lil_flags) {
3311 case LL_LEASE_RESYNC_DONE:
3313 OBD_FREE(data, data_size);
3315 case LL_LEASE_LAYOUT_MERGE:
3316 case LL_LEASE_LAYOUT_SPLIT:
3320 case LL_LEASE_PCC_ATTACH:
3323 rc = pcc_readwrite_attach_fini(file, inode,
3324 param.pa_layout_gen,
3331 rc = ll_lease_type_from_fmode(fmode);
3335 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3338 struct inode *inode = file_inode(file);
3339 struct ll_inode_info *lli = ll_i2info(inode);
3340 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3341 struct obd_client_handle *och = NULL;
3342 __u64 open_flags = 0;
3348 switch (ioc->lil_mode) {
3349 case LL_LEASE_WRLCK:
3350 if (!(file->f_mode & FMODE_WRITE))
3352 fmode = FMODE_WRITE;
3354 case LL_LEASE_RDLCK:
3355 if (!(file->f_mode & FMODE_READ))
3359 case LL_LEASE_UNLCK:
3360 RETURN(ll_file_unlock_lease(file, ioc, arg));
3365 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3367 /* apply for lease */
3368 if (ioc->lil_flags & LL_LEASE_RESYNC)
3369 open_flags = MDS_OPEN_RESYNC;
3370 och = ll_lease_open(inode, file, fmode, open_flags);
3372 RETURN(PTR_ERR(och));
3374 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3375 rc = ll_lease_file_resync(och, inode, arg);
3377 ll_lease_close(och, inode, NULL);
3380 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3382 ll_lease_close(och, inode, NULL);
3388 mutex_lock(&lli->lli_och_mutex);
3389 if (fd->fd_lease_och == NULL) {
3390 fd->fd_lease_och = och;
3393 mutex_unlock(&lli->lli_och_mutex);
3395 /* impossible now that only excl is supported for now */
3396 ll_lease_close(och, inode, &lease_broken);
3402 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3404 struct ll_inode_info *lli = ll_i2info(inode);
3405 struct ll_sb_info *sbi = ll_i2sbi(inode);
3406 __u64 now = ktime_get_real_seconds();
3409 spin_lock(&lli->lli_heat_lock);
3410 heat->lh_flags = lli->lli_heat_flags;
3411 for (i = 0; i < heat->lh_count; i++)
3412 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3413 now, sbi->ll_heat_decay_weight,
3414 sbi->ll_heat_period_second);
3415 spin_unlock(&lli->lli_heat_lock);
3418 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3420 struct ll_inode_info *lli = ll_i2info(inode);
3423 spin_lock(&lli->lli_heat_lock);
3424 if (flags & LU_HEAT_FLAG_CLEAR)
3425 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3427 if (flags & LU_HEAT_FLAG_OFF)
3428 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3430 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3432 spin_unlock(&lli->lli_heat_lock);
3438 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3440 struct inode *inode = file_inode(file);
3441 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
3445 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3446 PFID(ll_inode2fid(inode)), inode, cmd);
3447 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3449 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3450 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3454 case LL_IOC_GETFLAGS:
3455 /* Get the current value of the file flags */
3456 return put_user(fd->fd_flags, (int __user *)arg);
3457 case LL_IOC_SETFLAGS:
3458 case LL_IOC_CLRFLAGS:
3459 /* Set or clear specific file flags */
3460 /* XXX This probably needs checks to ensure the flags are
3461 * not abused, and to handle any flag side effects.
3463 if (get_user(flags, (int __user *) arg))
3466 if (cmd == LL_IOC_SETFLAGS) {
3467 if ((flags & LL_FILE_IGNORE_LOCK) &&
3468 !(file->f_flags & O_DIRECT)) {
3469 CERROR("%s: unable to disable locking on "
3470 "non-O_DIRECT file\n", current->comm);
3474 fd->fd_flags |= flags;
3476 fd->fd_flags &= ~flags;
3479 case LL_IOC_LOV_SETSTRIPE:
3480 case LL_IOC_LOV_SETSTRIPE_NEW:
3481 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3482 case LL_IOC_LOV_SETEA:
3483 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3484 case LL_IOC_LOV_SWAP_LAYOUTS: {
3486 struct lustre_swap_layouts lsl;
3488 if (copy_from_user(&lsl, (char __user *)arg,
3489 sizeof(struct lustre_swap_layouts)))
3492 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3495 file2 = fget(lsl.sl_fd);
3499 /* O_WRONLY or O_RDWR */
3500 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3501 GOTO(out, rc = -EPERM);
3503 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3504 struct inode *inode2;
3505 struct ll_inode_info *lli;
3506 struct obd_client_handle *och = NULL;
3508 lli = ll_i2info(inode);
3509 mutex_lock(&lli->lli_och_mutex);
3510 if (fd->fd_lease_och != NULL) {
3511 och = fd->fd_lease_och;
3512 fd->fd_lease_och = NULL;
3514 mutex_unlock(&lli->lli_och_mutex);
3516 GOTO(out, rc = -ENOLCK);
3517 inode2 = file_inode(file2);
3518 rc = ll_swap_layouts_close(och, inode, inode2);
3520 rc = ll_swap_layouts(file, file2, &lsl);
3526 case LL_IOC_LOV_GETSTRIPE:
3527 case LL_IOC_LOV_GETSTRIPE_NEW:
3528 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
3529 case FS_IOC_GETFLAGS:
3530 case FS_IOC_SETFLAGS:
3531 RETURN(ll_iocontrol(inode, file, cmd, arg));
3532 case FSFILT_IOC_GETVERSION:
3533 case FS_IOC_GETVERSION:
3534 RETURN(put_user(inode->i_generation, (int __user *)arg));
3535 /* We need to special case any other ioctls we want to handle,
3536 * to send them to the MDS/OST as appropriate and to properly
3537 * network encode the arg field. */
3538 case FS_IOC_SETVERSION:
3541 case LL_IOC_GROUP_LOCK:
3542 RETURN(ll_get_grouplock(inode, file, arg));
3543 case LL_IOC_GROUP_UNLOCK:
3544 RETURN(ll_put_grouplock(inode, file, arg));
3545 case IOC_OBD_STATFS:
3546 RETURN(ll_obd_statfs(inode, (void __user *)arg));
3548 case LL_IOC_FLUSHCTX:
3549 RETURN(ll_flush_ctx(inode));
3550 case LL_IOC_PATH2FID: {
3551 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
3552 sizeof(struct lu_fid)))
3557 case LL_IOC_GETPARENT:
3558 RETURN(ll_getparent(file, (struct getparent __user *)arg));
3560 case OBD_IOC_FID2PATH:
3561 RETURN(ll_fid2path(inode, (void __user *)arg));
3562 case LL_IOC_DATA_VERSION: {
3563 struct ioc_data_version idv;
3566 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
3569 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
3570 rc = ll_ioc_data_version(inode, &idv);
3573 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
3579 case LL_IOC_GET_MDTIDX: {
3582 mdtidx = ll_get_mdt_idx(inode);
3586 if (put_user((int)mdtidx, (int __user *)arg))
3591 case OBD_IOC_GETDTNAME:
3592 case OBD_IOC_GETMDNAME:
3593 RETURN(ll_get_obd_name(inode, cmd, arg));
3594 case LL_IOC_HSM_STATE_GET: {
3595 struct md_op_data *op_data;
3596 struct hsm_user_state *hus;
3603 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3604 LUSTRE_OPC_ANY, hus);
3605 if (IS_ERR(op_data)) {
3607 RETURN(PTR_ERR(op_data));
3610 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3613 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
3616 ll_finish_md_op_data(op_data);
3620 case LL_IOC_HSM_STATE_SET: {
3621 struct hsm_state_set *hss;
3628 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
3633 rc = ll_hsm_state_set(inode, hss);
3638 case LL_IOC_HSM_ACTION: {
3639 struct md_op_data *op_data;
3640 struct hsm_current_action *hca;
3647 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3648 LUSTRE_OPC_ANY, hca);
3649 if (IS_ERR(op_data)) {
3651 RETURN(PTR_ERR(op_data));
3654 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3657 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
3660 ll_finish_md_op_data(op_data);
3664 case LL_IOC_SET_LEASE_OLD: {
3665 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
3667 RETURN(ll_file_set_lease(file, &ioc, 0));
3669 case LL_IOC_SET_LEASE: {
3670 struct ll_ioc_lease ioc;
3672 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
3675 RETURN(ll_file_set_lease(file, &ioc, arg));
3677 case LL_IOC_GET_LEASE: {
3678 struct ll_inode_info *lli = ll_i2info(inode);
3679 struct ldlm_lock *lock = NULL;
3682 mutex_lock(&lli->lli_och_mutex);
3683 if (fd->fd_lease_och != NULL) {
3684 struct obd_client_handle *och = fd->fd_lease_och;
3686 lock = ldlm_handle2lock(&och->och_lease_handle);
3688 lock_res_and_lock(lock);
3689 if (!ldlm_is_cancel(lock))
3690 fmode = och->och_flags;
3692 unlock_res_and_lock(lock);
3693 LDLM_LOCK_PUT(lock);
3696 mutex_unlock(&lli->lli_och_mutex);
3698 RETURN(ll_lease_type_from_fmode(fmode));
3700 case LL_IOC_HSM_IMPORT: {
3701 struct hsm_user_import *hui;
3707 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
3712 rc = ll_hsm_import(inode, file, hui);
3717 case LL_IOC_FUTIMES_3: {
3718 struct ll_futimes_3 lfu;
3720 if (copy_from_user(&lfu,
3721 (const struct ll_futimes_3 __user *)arg,
3725 RETURN(ll_file_futimes_3(file, &lfu));
3727 case LL_IOC_LADVISE: {
3728 struct llapi_ladvise_hdr *k_ladvise_hdr;
3729 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
3732 int alloc_size = sizeof(*k_ladvise_hdr);
3735 u_ladvise_hdr = (void __user *)arg;
3736 OBD_ALLOC_PTR(k_ladvise_hdr);
3737 if (k_ladvise_hdr == NULL)
3740 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3741 GOTO(out_ladvise, rc = -EFAULT);
3743 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
3744 k_ladvise_hdr->lah_count < 1)
3745 GOTO(out_ladvise, rc = -EINVAL);
3747 num_advise = k_ladvise_hdr->lah_count;
3748 if (num_advise >= LAH_COUNT_MAX)
3749 GOTO(out_ladvise, rc = -EFBIG);
3751 OBD_FREE_PTR(k_ladvise_hdr);
3752 alloc_size = offsetof(typeof(*k_ladvise_hdr),
3753 lah_advise[num_advise]);
3754 OBD_ALLOC(k_ladvise_hdr, alloc_size);
3755 if (k_ladvise_hdr == NULL)
3759 * TODO: submit multiple advices to one server in a single RPC
3761 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3762 GOTO(out_ladvise, rc = -EFAULT);
3764 for (i = 0; i < num_advise; i++) {
3765 struct llapi_lu_ladvise *k_ladvise =
3766 &k_ladvise_hdr->lah_advise[i];
3767 struct llapi_lu_ladvise __user *u_ladvise =
3768 &u_ladvise_hdr->lah_advise[i];
3770 rc = ll_ladvise_sanity(inode, k_ladvise);
3772 GOTO(out_ladvise, rc);
3774 switch (k_ladvise->lla_advice) {
3775 case LU_LADVISE_LOCKNOEXPAND:
3776 rc = ll_lock_noexpand(file,
3777 k_ladvise->lla_peradvice_flags);
3778 GOTO(out_ladvise, rc);
3779 case LU_LADVISE_LOCKAHEAD:
3781 rc = ll_file_lock_ahead(file, k_ladvise);
3784 GOTO(out_ladvise, rc);
3787 &u_ladvise->lla_lockahead_result))
3788 GOTO(out_ladvise, rc = -EFAULT);
3791 rc = ll_ladvise(inode, file,
3792 k_ladvise_hdr->lah_flags,
3795 GOTO(out_ladvise, rc);
3802 OBD_FREE(k_ladvise_hdr, alloc_size);
3805 case LL_IOC_FLR_SET_MIRROR: {
3806 /* mirror I/O must be direct to avoid polluting page cache
3808 if (!(file->f_flags & O_DIRECT))
3811 fd->fd_designated_mirror = (__u32)arg;
3814 case LL_IOC_FSGETXATTR:
3815 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
3816 case LL_IOC_FSSETXATTR:
3817 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
3819 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
3820 case LL_IOC_HEAT_GET: {
3821 struct lu_heat uheat;
3822 struct lu_heat *heat;
3825 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
3828 if (uheat.lh_count > OBD_HEAT_COUNT)
3829 uheat.lh_count = OBD_HEAT_COUNT;
3831 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
3832 OBD_ALLOC(heat, size);
3836 heat->lh_count = uheat.lh_count;
3837 ll_heat_get(inode, heat);
3838 rc = copy_to_user((char __user *)arg, heat, size);
3839 OBD_FREE(heat, size);
3840 RETURN(rc ? -EFAULT : 0);
3842 case LL_IOC_HEAT_SET: {
3845 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
3848 rc = ll_heat_set(inode, flags);
3851 case LL_IOC_PCC_DETACH: {
3852 struct lu_pcc_detach *detach;
3854 OBD_ALLOC_PTR(detach);
3858 if (copy_from_user(detach,
3859 (const struct lu_pcc_detach __user *)arg,
3861 GOTO(out_detach_free, rc = -EFAULT);
3863 if (!S_ISREG(inode->i_mode))
3864 GOTO(out_detach_free, rc = -EINVAL);
3866 if (!inode_owner_or_capable(inode))
3867 GOTO(out_detach_free, rc = -EPERM);
3869 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
3871 OBD_FREE_PTR(detach);
3874 case LL_IOC_PCC_STATE: {
3875 struct lu_pcc_state __user *ustate =
3876 (struct lu_pcc_state __user *)arg;
3877 struct lu_pcc_state *state;
3879 OBD_ALLOC_PTR(state);
3883 if (copy_from_user(state, ustate, sizeof(*state)))
3884 GOTO(out_state, rc = -EFAULT);
3886 rc = pcc_ioctl_state(file, inode, state);
3888 GOTO(out_state, rc);
3890 if (copy_to_user(ustate, state, sizeof(*state)))
3891 GOTO(out_state, rc = -EFAULT);
3894 OBD_FREE_PTR(state);
3898 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
3899 (void __user *)arg));
3903 #ifndef HAVE_FILE_LLSEEK_SIZE
3904 static inline loff_t
3905 llseek_execute(struct file *file, loff_t offset, loff_t maxsize)
3907 if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
3909 if (offset > maxsize)
3912 if (offset != file->f_pos) {
3913 file->f_pos = offset;
3914 file->f_version = 0;
3920 generic_file_llseek_size(struct file *file, loff_t offset, int origin,
3921 loff_t maxsize, loff_t eof)
3923 struct inode *inode = file_inode(file);
3931 * Here we special-case the lseek(fd, 0, SEEK_CUR)
3932 * position-querying operation. Avoid rewriting the "same"
3933 * f_pos value back to the file because a concurrent read(),
3934 * write() or lseek() might have altered it
3939 * f_lock protects against read/modify/write race with other
3940 * SEEK_CURs. Note that parallel writes and reads behave
3944 offset = llseek_execute(file, file->f_pos + offset, maxsize);
3945 inode_unlock(inode);
3949 * In the generic case the entire file is data, so as long as
3950 * offset isn't at the end of the file then the offset is data.
3957 * There is a virtual hole at the end of the file, so as long as
3958 * offset isn't i_size or larger, return i_size.
3966 return llseek_execute(file, offset, maxsize);
3970 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
3972 struct inode *inode = file_inode(file);
3973 loff_t retval, eof = 0;
3976 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
3977 (origin == SEEK_CUR) ? file->f_pos : 0);
3978 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
3979 PFID(ll_inode2fid(inode)), inode, retval, retval,
3981 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
3983 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
3984 retval = ll_glimpse_size(inode);
3987 eof = i_size_read(inode);
3990 retval = ll_generic_file_llseek_size(file, offset, origin,
3991 ll_file_maxbytes(inode), eof);
3995 static int ll_flush(struct file *file, fl_owner_t id)
3997 struct inode *inode = file_inode(file);
3998 struct ll_inode_info *lli = ll_i2info(inode);
3999 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
4002 LASSERT(!S_ISDIR(inode->i_mode));
4004 /* catch async errors that were recorded back when async writeback
4005 * failed for pages in this mapping. */
4006 rc = lli->lli_async_rc;
4007 lli->lli_async_rc = 0;
4008 if (lli->lli_clob != NULL) {
4009 err = lov_read_and_clear_async_rc(lli->lli_clob);
4014 /* The application has been told write failure already.
4015 * Do not report failure again. */
4016 if (fd->fd_write_failed)
4018 return rc ? -EIO : 0;
4022 * Called to make sure a portion of file has been written out.
4023 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4025 * Return how many pages have been written.
4027 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4028 enum cl_fsync_mode mode, int ignore_layout)
4032 struct cl_fsync_io *fio;
4037 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4038 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4041 env = cl_env_get(&refcheck);
4043 RETURN(PTR_ERR(env));
4045 io = vvp_env_thread_io(env);
4046 io->ci_obj = ll_i2info(inode)->lli_clob;
4047 io->ci_ignore_layout = ignore_layout;
4049 /* initialize parameters for sync */
4050 fio = &io->u.ci_fsync;
4051 fio->fi_start = start;
4053 fio->fi_fid = ll_inode2fid(inode);
4054 fio->fi_mode = mode;
4055 fio->fi_nr_written = 0;
4057 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4058 result = cl_io_loop(env, io);
4060 result = io->ci_result;
4062 result = fio->fi_nr_written;
4063 cl_io_fini(env, io);
4064 cl_env_put(env, &refcheck);
4070 * When dentry is provided (the 'else' case), file_dentry() may be
4071 * null and dentry must be used directly rather than pulled from
4072 * file_dentry() as is done otherwise.
4075 #ifdef HAVE_FILE_FSYNC_4ARGS
4076 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4078 struct dentry *dentry = file_dentry(file);
4079 #elif defined(HAVE_FILE_FSYNC_2ARGS)
4080 int ll_fsync(struct file *file, int datasync)
4082 struct dentry *dentry = file_dentry(file);
4084 loff_t end = LLONG_MAX;
4086 int ll_fsync(struct file *file, struct dentry *dentry, int datasync)
4089 loff_t end = LLONG_MAX;
4091 struct inode *inode = dentry->d_inode;
4092 struct ll_inode_info *lli = ll_i2info(inode);
4093 struct ptlrpc_request *req;
4098 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
4099 PFID(ll_inode2fid(inode)), inode);
4100 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
4102 #ifdef HAVE_FILE_FSYNC_4ARGS
4103 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4106 /* fsync's caller has already called _fdata{sync,write}, we want
4107 * that IO to finish before calling the osc and mdc sync methods */
4108 rc = filemap_fdatawait(inode->i_mapping);
4111 /* catch async errors that were recorded back when async writeback
4112 * failed for pages in this mapping. */
4113 if (!S_ISDIR(inode->i_mode)) {
4114 err = lli->lli_async_rc;
4115 lli->lli_async_rc = 0;
4118 if (lli->lli_clob != NULL) {
4119 err = lov_read_and_clear_async_rc(lli->lli_clob);
4125 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4129 ptlrpc_req_finished(req);
4131 if (S_ISREG(inode->i_mode)) {
4132 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
4135 /* Sync metadata on MDT first, and then sync the cached data
4138 err = pcc_fsync(file, start, end, datasync, &cached);
4140 err = cl_sync_file_range(inode, start, end,
4142 if (rc == 0 && err < 0)
4145 fd->fd_write_failed = true;
4147 fd->fd_write_failed = false;
4150 #ifdef HAVE_FILE_FSYNC_4ARGS
4151 inode_unlock(inode);
4157 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4159 struct inode *inode = file_inode(file);
4160 struct ll_sb_info *sbi = ll_i2sbi(inode);
4161 struct ldlm_enqueue_info einfo = {
4162 .ei_type = LDLM_FLOCK,
4163 .ei_cb_cp = ldlm_flock_completion_ast,
4164 .ei_cbdata = file_lock,
4166 struct md_op_data *op_data;
4167 struct lustre_handle lockh = { 0 };
4168 union ldlm_policy_data flock = { { 0 } };
4169 int fl_type = file_lock->fl_type;
4175 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4176 PFID(ll_inode2fid(inode)), file_lock);
4178 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
4180 if (file_lock->fl_flags & FL_FLOCK) {
4181 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4182 /* flocks are whole-file locks */
4183 flock.l_flock.end = OFFSET_MAX;
4184 /* For flocks owner is determined by the local file desctiptor*/
4185 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4186 } else if (file_lock->fl_flags & FL_POSIX) {
4187 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4188 flock.l_flock.start = file_lock->fl_start;
4189 flock.l_flock.end = file_lock->fl_end;
4193 flock.l_flock.pid = file_lock->fl_pid;
4195 /* Somewhat ugly workaround for svc lockd.
4196 * lockd installs custom fl_lmops->lm_compare_owner that checks
4197 * for the fl_owner to be the same (which it always is on local node
4198 * I guess between lockd processes) and then compares pid.
4199 * As such we assign pid to the owner field to make it all work,
4200 * conflict with normal locks is unlikely since pid space and
4201 * pointer space for current->files are not intersecting */
4202 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4203 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4207 einfo.ei_mode = LCK_PR;
4210 /* An unlock request may or may not have any relation to
4211 * existing locks so we may not be able to pass a lock handle
4212 * via a normal ldlm_lock_cancel() request. The request may even
4213 * unlock a byte range in the middle of an existing lock. In
4214 * order to process an unlock request we need all of the same
4215 * information that is given with a normal read or write record
4216 * lock request. To avoid creating another ldlm unlock (cancel)
4217 * message we'll treat a LCK_NL flock request as an unlock. */
4218 einfo.ei_mode = LCK_NL;
4221 einfo.ei_mode = LCK_PW;
4224 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4239 flags = LDLM_FL_BLOCK_NOWAIT;
4245 flags = LDLM_FL_TEST_LOCK;
4248 CERROR("unknown fcntl lock command: %d\n", cmd);
4252 /* Save the old mode so that if the mode in the lock changes we
4253 * can decrement the appropriate reader or writer refcount. */
4254 file_lock->fl_type = einfo.ei_mode;
4256 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4257 LUSTRE_OPC_ANY, NULL);
4258 if (IS_ERR(op_data))
4259 RETURN(PTR_ERR(op_data));
4261 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4262 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4263 flock.l_flock.pid, flags, einfo.ei_mode,
4264 flock.l_flock.start, flock.l_flock.end);
4266 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4269 /* Restore the file lock type if not TEST lock. */
4270 if (!(flags & LDLM_FL_TEST_LOCK))
4271 file_lock->fl_type = fl_type;
4273 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4274 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4275 !(flags & LDLM_FL_TEST_LOCK))
4276 rc2 = locks_lock_file_wait(file, file_lock);
4278 if ((file_lock->fl_flags & FL_FLOCK) &&
4279 (rc == 0 || file_lock->fl_type == F_UNLCK))
4280 rc2 = flock_lock_file_wait(file, file_lock);
4281 if ((file_lock->fl_flags & FL_POSIX) &&
4282 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4283 !(flags & LDLM_FL_TEST_LOCK))
4284 rc2 = posix_lock_file_wait(file, file_lock);
4285 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4287 if (rc2 && file_lock->fl_type != F_UNLCK) {
4288 einfo.ei_mode = LCK_NL;
4289 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4294 ll_finish_md_op_data(op_data);
4299 int ll_get_fid_by_name(struct inode *parent, const char *name,
4300 int namelen, struct lu_fid *fid,
4301 struct inode **inode)
4303 struct md_op_data *op_data = NULL;
4304 struct mdt_body *body;
4305 struct ptlrpc_request *req;
4309 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4310 LUSTRE_OPC_ANY, NULL);
4311 if (IS_ERR(op_data))
4312 RETURN(PTR_ERR(op_data));
4314 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4315 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4316 ll_finish_md_op_data(op_data);
4320 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4322 GOTO(out_req, rc = -EFAULT);
4324 *fid = body->mbo_fid1;
4327 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
4329 ptlrpc_req_finished(req);
4333 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4336 struct dentry *dchild = NULL;
4337 struct inode *child_inode = NULL;
4338 struct md_op_data *op_data;
4339 struct ptlrpc_request *request = NULL;
4340 struct obd_client_handle *och = NULL;
4342 struct mdt_body *body;
4343 __u64 data_version = 0;
4344 size_t namelen = strlen(name);
4345 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4349 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4350 PFID(ll_inode2fid(parent)), name,
4351 lum->lum_stripe_offset, lum->lum_stripe_count);
4353 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4354 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4355 lustre_swab_lmv_user_md(lum);
4357 /* Get child FID first */
4358 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4361 dchild = d_lookup(file_dentry(file), &qstr);
4363 if (dchild->d_inode)
4364 child_inode = igrab(dchild->d_inode);
4369 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4378 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4379 OBD_CONNECT2_DIR_MIGRATE)) {
4380 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4381 ll_dir_striped(child_inode)) {
4382 CERROR("%s: MDT doesn't support stripe directory "
4383 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4384 GOTO(out_iput, rc = -EOPNOTSUPP);
4389 * lfs migrate command needs to be blocked on the client
4390 * by checking the migrate FID against the FID of the
4393 if (child_inode == parent->i_sb->s_root->d_inode)
4394 GOTO(out_iput, rc = -EINVAL);
4396 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4397 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
4398 if (IS_ERR(op_data))
4399 GOTO(out_iput, rc = PTR_ERR(op_data));
4401 inode_lock(child_inode);
4402 op_data->op_fid3 = *ll_inode2fid(child_inode);
4403 if (!fid_is_sane(&op_data->op_fid3)) {
4404 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4405 ll_i2sbi(parent)->ll_fsname, name,
4406 PFID(&op_data->op_fid3));
4407 GOTO(out_unlock, rc = -EINVAL);
4410 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4411 op_data->op_data = lum;
4412 op_data->op_data_size = lumlen;
4415 if (S_ISREG(child_inode->i_mode)) {
4416 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4420 GOTO(out_unlock, rc);
4423 rc = ll_data_version(child_inode, &data_version,
4426 GOTO(out_close, rc);
4428 op_data->op_open_handle = och->och_open_handle;
4429 op_data->op_data_version = data_version;
4430 op_data->op_lease_handle = och->och_lease_handle;
4431 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4433 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4434 och->och_mod->mod_open_req->rq_replay = 0;
4435 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4438 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name, namelen,
4439 name, namelen, &request);
4441 LASSERT(request != NULL);
4442 ll_update_times(request, parent);
4445 if (rc == 0 || rc == -EAGAIN) {
4446 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4447 LASSERT(body != NULL);
4449 /* If the server does release layout lock, then we cleanup
4450 * the client och here, otherwise release it in out_close: */
4451 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4452 obd_mod_put(och->och_mod);
4453 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4455 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4461 if (request != NULL) {
4462 ptlrpc_req_finished(request);
4466 /* Try again if the lease has cancelled. */
4467 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4472 ll_lease_close(och, child_inode, NULL);
4474 clear_nlink(child_inode);
4476 inode_unlock(child_inode);
4477 ll_finish_md_op_data(op_data);
4484 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
4492 * test if some locks matching bits and l_req_mode are acquired
4493 * - bits can be in different locks
4494 * - if found clear the common lock bits in *bits
4495 * - the bits not found, are kept in *bits
4497 * \param bits [IN] searched lock bits [IN]
4498 * \param l_req_mode [IN] searched lock mode
4499 * \retval boolean, true iff all bits are found
4501 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
4503 struct lustre_handle lockh;
4504 union ldlm_policy_data policy;
4505 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
4506 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
4515 fid = &ll_i2info(inode)->lli_fid;
4516 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
4517 ldlm_lockname[mode]);
4519 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
4520 for (i = 0; i <= MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) {
4521 policy.l_inodebits.bits = *bits & (1 << i);
4522 if (policy.l_inodebits.bits == 0)
4525 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
4526 &policy, mode, &lockh)) {
4527 struct ldlm_lock *lock;
4529 lock = ldlm_handle2lock(&lockh);
4532 ~(lock->l_policy_data.l_inodebits.bits);
4533 LDLM_LOCK_PUT(lock);
4535 *bits &= ~policy.l_inodebits.bits;
4542 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
4543 struct lustre_handle *lockh, __u64 flags,
4544 enum ldlm_mode mode)
4546 union ldlm_policy_data policy = { .l_inodebits = { bits } };
4551 fid = &ll_i2info(inode)->lli_fid;
4552 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
4554 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
4555 fid, LDLM_IBITS, &policy, mode, lockh);
4560 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
4562 /* Already unlinked. Just update nlink and return success */
4563 if (rc == -ENOENT) {
4565 /* If it is striped directory, and there is bad stripe
4566 * Let's revalidate the dentry again, instead of returning
4568 if (ll_dir_striped(inode))
4571 /* This path cannot be hit for regular files unless in
4572 * case of obscure races, so no need to to validate
4574 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
4576 } else if (rc != 0) {
4577 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
4578 "%s: revalidate FID "DFID" error: rc = %d\n",
4579 ll_i2sbi(inode)->ll_fsname,
4580 PFID(ll_inode2fid(inode)), rc);
4586 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
4588 struct inode *inode = dentry->d_inode;
4589 struct obd_export *exp = ll_i2mdexp(inode);
4590 struct lookup_intent oit = {
4593 struct ptlrpc_request *req = NULL;
4594 struct md_op_data *op_data;
4598 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
4599 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
4601 /* Call getattr by fid, so do not provide name at all. */
4602 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
4603 LUSTRE_OPC_ANY, NULL);
4604 if (IS_ERR(op_data))
4605 RETURN(PTR_ERR(op_data));
4607 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
4608 ll_finish_md_op_data(op_data);
4610 rc = ll_inode_revalidate_fini(inode, rc);
4614 rc = ll_revalidate_it_finish(req, &oit, dentry);
4616 ll_intent_release(&oit);
4620 /* Unlinked? Unhash dentry, so it is not picked up later by
4621 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
4622 * here to preserve get_cwd functionality on 2.6.
4624 if (!dentry->d_inode->i_nlink) {
4625 ll_lock_dcache(inode);
4626 d_lustre_invalidate(dentry, 0);
4627 ll_unlock_dcache(inode);
4630 ll_lookup_finish_locks(&oit, dentry);
4632 ptlrpc_req_finished(req);
4637 static int ll_merge_md_attr(struct inode *inode)
4639 struct ll_inode_info *lli = ll_i2info(inode);
4640 struct cl_attr attr = { 0 };
4643 LASSERT(lli->lli_lsm_md != NULL);
4645 if (!lmv_dir_striped(lli->lli_lsm_md))
4648 down_read(&lli->lli_lsm_sem);
4649 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
4650 &attr, ll_md_blocking_ast);
4651 up_read(&lli->lli_lsm_sem);
4655 set_nlink(inode, attr.cat_nlink);
4656 inode->i_blocks = attr.cat_blocks;
4657 i_size_write(inode, attr.cat_size);
4659 ll_i2info(inode)->lli_atime = attr.cat_atime;
4660 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
4661 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
4666 int ll_getattr_dentry(struct dentry *de, struct kstat *stat)
4668 struct inode *inode = de->d_inode;
4669 struct ll_sb_info *sbi = ll_i2sbi(inode);
4670 struct ll_inode_info *lli = ll_i2info(inode);
4673 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1);
4675 rc = ll_inode_revalidate(de, IT_GETATTR);
4679 if (S_ISREG(inode->i_mode)) {
4682 rc = pcc_inode_getattr(inode, &cached);
4683 if (cached && rc < 0)
4686 /* In case of restore, the MDT has the right size and has
4687 * already send it back without granting the layout lock,
4688 * inode is up-to-date so glimpse is useless.
4689 * Also to glimpse we need the layout, in case of a running
4690 * restore the MDT holds the layout lock so the glimpse will
4691 * block up to the end of restore (getattr will block)
4693 if (!cached && !ll_file_test_flag(lli, LLIF_FILE_RESTORING)) {
4694 rc = ll_glimpse_size(inode);
4699 /* If object isn't regular a file then don't validate size. */
4700 if (ll_dir_striped(inode)) {
4701 rc = ll_merge_md_attr(inode);
4706 inode->i_atime.tv_sec = lli->lli_atime;
4707 inode->i_mtime.tv_sec = lli->lli_mtime;
4708 inode->i_ctime.tv_sec = lli->lli_ctime;
4711 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
4713 if (ll_need_32bit_api(sbi)) {
4714 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
4715 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
4716 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
4718 stat->ino = inode->i_ino;
4719 stat->dev = inode->i_sb->s_dev;
4720 stat->rdev = inode->i_rdev;
4723 stat->mode = inode->i_mode;
4724 stat->uid = inode->i_uid;
4725 stat->gid = inode->i_gid;
4726 stat->atime = inode->i_atime;
4727 stat->mtime = inode->i_mtime;
4728 stat->ctime = inode->i_ctime;
4729 stat->blksize = sbi->ll_stat_blksize ?: 1 << inode->i_blkbits;
4731 stat->nlink = inode->i_nlink;
4732 stat->size = i_size_read(inode);
4733 stat->blocks = inode->i_blocks;
4738 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
4739 int ll_getattr(const struct path *path, struct kstat *stat,
4740 u32 request_mask, unsigned int flags)
4742 struct dentry *de = path->dentry;
4744 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
4747 return ll_getattr_dentry(de, stat);
4750 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4751 __u64 start, __u64 len)
4755 struct fiemap *fiemap;
4756 unsigned int extent_count = fieinfo->fi_extents_max;
4758 num_bytes = sizeof(*fiemap) + (extent_count *
4759 sizeof(struct fiemap_extent));
4760 OBD_ALLOC_LARGE(fiemap, num_bytes);
4765 fiemap->fm_flags = fieinfo->fi_flags;
4766 fiemap->fm_extent_count = fieinfo->fi_extents_max;
4767 fiemap->fm_start = start;
4768 fiemap->fm_length = len;
4769 if (extent_count > 0 &&
4770 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
4771 sizeof(struct fiemap_extent)) != 0)
4772 GOTO(out, rc = -EFAULT);
4774 rc = ll_do_fiemap(inode, fiemap, num_bytes);
4776 fieinfo->fi_flags = fiemap->fm_flags;
4777 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
4778 if (extent_count > 0 &&
4779 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
4780 fiemap->fm_mapped_extents *
4781 sizeof(struct fiemap_extent)) != 0)
4782 GOTO(out, rc = -EFAULT);
4784 OBD_FREE_LARGE(fiemap, num_bytes);
4788 struct posix_acl *ll_get_acl(struct inode *inode, int type)
4790 struct ll_inode_info *lli = ll_i2info(inode);
4791 struct posix_acl *acl = NULL;
4794 spin_lock(&lli->lli_lock);
4795 /* VFS' acl_permission_check->check_acl will release the refcount */
4796 acl = posix_acl_dup(lli->lli_posix_acl);
4797 spin_unlock(&lli->lli_lock);
4802 #ifdef HAVE_IOP_SET_ACL
4803 #ifdef CONFIG_FS_POSIX_ACL
4804 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type)
4806 struct ll_sb_info *sbi = ll_i2sbi(inode);
4807 struct ptlrpc_request *req = NULL;
4808 const char *name = NULL;
4810 size_t value_size = 0;
4815 case ACL_TYPE_ACCESS:
4816 name = XATTR_NAME_POSIX_ACL_ACCESS;
4818 rc = posix_acl_update_mode(inode, &inode->i_mode, &acl);
4821 case ACL_TYPE_DEFAULT:
4822 name = XATTR_NAME_POSIX_ACL_DEFAULT;
4823 if (!S_ISDIR(inode->i_mode))
4824 rc = acl ? -EACCES : 0;
4835 value_size = posix_acl_xattr_size(acl->a_count);
4836 value = kmalloc(value_size, GFP_NOFS);
4838 GOTO(out, rc = -ENOMEM);
4840 rc = posix_acl_to_xattr(&init_user_ns, acl, value, value_size);
4842 GOTO(out_value, rc);
4845 rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode),
4846 value ? OBD_MD_FLXATTR : OBD_MD_FLXATTRRM,
4847 name, value, value_size, 0, 0, &req);
4849 ptlrpc_req_finished(req);
4854 forget_cached_acl(inode, type);
4856 set_cached_acl(inode, type, acl);
4859 #endif /* CONFIG_FS_POSIX_ACL */
4860 #endif /* HAVE_IOP_SET_ACL */
4862 #ifndef HAVE_GENERIC_PERMISSION_2ARGS
4864 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
4865 ll_check_acl(struct inode *inode, int mask, unsigned int flags)
4867 ll_check_acl(struct inode *inode, int mask)
4870 # ifdef CONFIG_FS_POSIX_ACL
4871 struct posix_acl *acl;
4875 # ifdef HAVE_GENERIC_PERMISSION_4ARGS
4876 if (flags & IPERM_FLAG_RCU)
4879 acl = ll_get_acl(inode, ACL_TYPE_ACCESS);
4884 rc = posix_acl_permission(inode, acl, mask);
4885 posix_acl_release(acl);
4888 # else /* !CONFIG_FS_POSIX_ACL */
4890 # endif /* CONFIG_FS_POSIX_ACL */
4892 #endif /* HAVE_GENERIC_PERMISSION_2ARGS */
4894 #ifdef HAVE_GENERIC_PERMISSION_4ARGS
4895 int ll_inode_permission(struct inode *inode, int mask, unsigned int flags)
4897 # ifdef HAVE_INODE_PERMISION_2ARGS
4898 int ll_inode_permission(struct inode *inode, int mask)
4900 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
4905 struct ll_sb_info *sbi;
4906 struct root_squash_info *squash;
4907 struct cred *cred = NULL;
4908 const struct cred *old_cred = NULL;
4910 bool squash_id = false;
4913 #ifdef MAY_NOT_BLOCK
4914 if (mask & MAY_NOT_BLOCK)
4916 #elif defined(HAVE_GENERIC_PERMISSION_4ARGS)
4917 if (flags & IPERM_FLAG_RCU)
4921 /* as root inode are NOT getting validated in lookup operation,
4922 * need to do it before permission check. */
4924 if (inode == inode->i_sb->s_root->d_inode) {
4925 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
4930 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
4931 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
4933 /* squash fsuid/fsgid if needed */
4934 sbi = ll_i2sbi(inode);
4935 squash = &sbi->ll_squash;
4936 if (unlikely(squash->rsi_uid != 0 &&
4937 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
4938 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
4942 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
4943 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
4944 squash->rsi_uid, squash->rsi_gid);
4946 /* update current process's credentials
4947 * and FS capability */
4948 cred = prepare_creds();
4952 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
4953 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
4954 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
4955 if ((1 << cap) & CFS_CAP_FS_MASK)
4956 cap_lower(cred->cap_effective, cap);
4958 old_cred = override_creds(cred);
4961 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM, 1);
4962 rc = ll_generic_permission(inode, mask, flags, ll_check_acl);
4963 /* restore current process's credentials and FS capability */
4965 revert_creds(old_cred);
4972 /* -o localflock - only provides locally consistent flock locks */
4973 struct file_operations ll_file_operations = {
4974 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
4975 # ifdef HAVE_SYNC_READ_WRITE
4976 .read = new_sync_read,
4977 .write = new_sync_write,
4979 .read_iter = ll_file_read_iter,
4980 .write_iter = ll_file_write_iter,
4981 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
4982 .read = ll_file_read,
4983 .aio_read = ll_file_aio_read,
4984 .write = ll_file_write,
4985 .aio_write = ll_file_aio_write,
4986 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
4987 .unlocked_ioctl = ll_file_ioctl,
4988 .open = ll_file_open,
4989 .release = ll_file_release,
4990 .mmap = ll_file_mmap,
4991 .llseek = ll_file_seek,
4992 .splice_read = ll_file_splice_read,
4997 struct file_operations ll_file_operations_flock = {
4998 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
4999 # ifdef HAVE_SYNC_READ_WRITE
5000 .read = new_sync_read,
5001 .write = new_sync_write,
5002 # endif /* HAVE_SYNC_READ_WRITE */
5003 .read_iter = ll_file_read_iter,
5004 .write_iter = ll_file_write_iter,
5005 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5006 .read = ll_file_read,
5007 .aio_read = ll_file_aio_read,
5008 .write = ll_file_write,
5009 .aio_write = ll_file_aio_write,
5010 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5011 .unlocked_ioctl = ll_file_ioctl,
5012 .open = ll_file_open,
5013 .release = ll_file_release,
5014 .mmap = ll_file_mmap,
5015 .llseek = ll_file_seek,
5016 .splice_read = ll_file_splice_read,
5019 .flock = ll_file_flock,
5020 .lock = ll_file_flock
5023 /* These are for -o noflock - to return ENOSYS on flock calls */
5024 struct file_operations ll_file_operations_noflock = {
5025 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5026 # ifdef HAVE_SYNC_READ_WRITE
5027 .read = new_sync_read,
5028 .write = new_sync_write,
5029 # endif /* HAVE_SYNC_READ_WRITE */
5030 .read_iter = ll_file_read_iter,
5031 .write_iter = ll_file_write_iter,
5032 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5033 .read = ll_file_read,
5034 .aio_read = ll_file_aio_read,
5035 .write = ll_file_write,
5036 .aio_write = ll_file_aio_write,
5037 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5038 .unlocked_ioctl = ll_file_ioctl,
5039 .open = ll_file_open,
5040 .release = ll_file_release,
5041 .mmap = ll_file_mmap,
5042 .llseek = ll_file_seek,
5043 .splice_read = ll_file_splice_read,
5046 .flock = ll_file_noflock,
5047 .lock = ll_file_noflock
5050 struct inode_operations ll_file_inode_operations = {
5051 .setattr = ll_setattr,
5052 .getattr = ll_getattr,
5053 .permission = ll_inode_permission,
5054 #ifdef HAVE_IOP_XATTR
5055 .setxattr = ll_setxattr,
5056 .getxattr = ll_getxattr,
5057 .removexattr = ll_removexattr,
5059 .listxattr = ll_listxattr,
5060 .fiemap = ll_fiemap,
5061 #ifdef HAVE_IOP_GET_ACL
5062 .get_acl = ll_get_acl,
5064 #ifdef HAVE_IOP_SET_ACL
5065 .set_acl = ll_set_acl,
5069 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5071 struct ll_inode_info *lli = ll_i2info(inode);
5072 struct cl_object *obj = lli->lli_clob;
5081 env = cl_env_get(&refcheck);
5083 RETURN(PTR_ERR(env));
5085 rc = cl_conf_set(env, lli->lli_clob, conf);
5089 if (conf->coc_opc == OBJECT_CONF_SET) {
5090 struct ldlm_lock *lock = conf->coc_lock;
5091 struct cl_layout cl = {
5095 LASSERT(lock != NULL);
5096 LASSERT(ldlm_has_layout(lock));
5098 /* it can only be allowed to match after layout is
5099 * applied to inode otherwise false layout would be
5100 * seen. Applying layout shoud happen before dropping
5101 * the intent lock. */
5102 ldlm_lock_allow_match(lock);
5104 rc = cl_object_layout_get(env, obj, &cl);
5109 DFID": layout version change: %u -> %u\n",
5110 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5112 ll_layout_version_set(lli, cl.cl_layout_gen);
5116 cl_env_put(env, &refcheck);
5121 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5122 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5125 struct ll_sb_info *sbi = ll_i2sbi(inode);
5126 struct ptlrpc_request *req;
5133 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5134 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5135 lock->l_lvb_data, lock->l_lvb_len);
5137 if (lock->l_lvb_data != NULL)
5140 /* if layout lock was granted right away, the layout is returned
5141 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5142 * blocked and then granted via completion ast, we have to fetch
5143 * layout here. Please note that we can't use the LVB buffer in
5144 * completion AST because it doesn't have a large enough buffer */
5145 rc = ll_get_default_mdsize(sbi, &lmmsize);
5149 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5150 XATTR_NAME_LOV, lmmsize, &req);
5153 GOTO(out, rc = 0); /* empty layout */
5160 if (lmmsize == 0) /* empty layout */
5163 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5165 GOTO(out, rc = -EFAULT);
5167 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5168 if (lvbdata == NULL)
5169 GOTO(out, rc = -ENOMEM);
5171 memcpy(lvbdata, lmm, lmmsize);
5172 lock_res_and_lock(lock);
5173 if (unlikely(lock->l_lvb_data == NULL)) {
5174 lock->l_lvb_type = LVB_T_LAYOUT;
5175 lock->l_lvb_data = lvbdata;
5176 lock->l_lvb_len = lmmsize;
5179 unlock_res_and_lock(lock);
5182 OBD_FREE_LARGE(lvbdata, lmmsize);
5187 ptlrpc_req_finished(req);
5192 * Apply the layout to the inode. Layout lock is held and will be released
5195 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5196 struct inode *inode)
5198 struct ll_inode_info *lli = ll_i2info(inode);
5199 struct ll_sb_info *sbi = ll_i2sbi(inode);
5200 struct ldlm_lock *lock;
5201 struct cl_object_conf conf;
5204 bool wait_layout = false;
5207 LASSERT(lustre_handle_is_used(lockh));
5209 lock = ldlm_handle2lock(lockh);
5210 LASSERT(lock != NULL);
5211 LASSERT(ldlm_has_layout(lock));
5213 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5214 PFID(&lli->lli_fid), inode);
5216 /* in case this is a caching lock and reinstate with new inode */
5217 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5219 lock_res_and_lock(lock);
5220 lvb_ready = ldlm_is_lvb_ready(lock);
5221 unlock_res_and_lock(lock);
5223 /* checking lvb_ready is racy but this is okay. The worst case is
5224 * that multi processes may configure the file on the same time. */
5228 rc = ll_layout_fetch(inode, lock);
5232 /* for layout lock, lmm is stored in lock's lvb.
5233 * lvb_data is immutable if the lock is held so it's safe to access it
5236 * set layout to file. Unlikely this will fail as old layout was
5237 * surely eliminated */
5238 memset(&conf, 0, sizeof conf);
5239 conf.coc_opc = OBJECT_CONF_SET;
5240 conf.coc_inode = inode;
5241 conf.coc_lock = lock;
5242 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5243 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5244 rc = ll_layout_conf(inode, &conf);
5246 /* refresh layout failed, need to wait */
5247 wait_layout = rc == -EBUSY;
5250 LDLM_LOCK_PUT(lock);
5251 ldlm_lock_decref(lockh, mode);
5253 /* wait for IO to complete if it's still being used. */
5255 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5256 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5258 memset(&conf, 0, sizeof conf);
5259 conf.coc_opc = OBJECT_CONF_WAIT;
5260 conf.coc_inode = inode;
5261 rc = ll_layout_conf(inode, &conf);
5265 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5266 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5272 * Issue layout intent RPC to MDS.
5273 * \param inode [in] file inode
5274 * \param intent [in] layout intent
5276 * \retval 0 on success
5277 * \retval < 0 error code
5279 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5281 struct ll_inode_info *lli = ll_i2info(inode);
5282 struct ll_sb_info *sbi = ll_i2sbi(inode);
5283 struct md_op_data *op_data;
5284 struct lookup_intent it;
5285 struct ptlrpc_request *req;
5289 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5290 0, 0, LUSTRE_OPC_ANY, NULL);
5291 if (IS_ERR(op_data))
5292 RETURN(PTR_ERR(op_data));
5294 op_data->op_data = intent;
5295 op_data->op_data_size = sizeof(*intent);
5297 memset(&it, 0, sizeof(it));
5298 it.it_op = IT_LAYOUT;
5299 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5300 intent->li_opc == LAYOUT_INTENT_TRUNC)
5301 it.it_flags = FMODE_WRITE;
5303 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5304 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5306 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5307 &ll_md_blocking_ast, 0);
5308 if (it.it_request != NULL)
5309 ptlrpc_req_finished(it.it_request);
5310 it.it_request = NULL;
5312 ll_finish_md_op_data(op_data);
5314 /* set lock data in case this is a new lock */
5316 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5318 ll_intent_drop_lock(&it);
5324 * This function checks if there exists a LAYOUT lock on the client side,
5325 * or enqueues it if it doesn't have one in cache.
5327 * This function will not hold layout lock so it may be revoked any time after
5328 * this function returns. Any operations depend on layout should be redone
5331 * This function should be called before lov_io_init() to get an uptodate
5332 * layout version, the caller should save the version number and after IO
5333 * is finished, this function should be called again to verify that layout
5334 * is not changed during IO time.
5336 int ll_layout_refresh(struct inode *inode, __u32 *gen)
5338 struct ll_inode_info *lli = ll_i2info(inode);
5339 struct ll_sb_info *sbi = ll_i2sbi(inode);
5340 struct lustre_handle lockh;
5341 struct layout_intent intent = {
5342 .li_opc = LAYOUT_INTENT_ACCESS,
5344 enum ldlm_mode mode;
5348 *gen = ll_layout_version_get(lli);
5349 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
5353 LASSERT(fid_is_sane(ll_inode2fid(inode)));
5354 LASSERT(S_ISREG(inode->i_mode));
5356 /* take layout lock mutex to enqueue layout lock exclusively. */
5357 mutex_lock(&lli->lli_layout_mutex);
5360 /* mostly layout lock is caching on the local side, so try to
5361 * match it before grabbing layout lock mutex. */
5362 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
5363 LCK_CR | LCK_CW | LCK_PR | LCK_PW);
5364 if (mode != 0) { /* hit cached lock */
5365 rc = ll_layout_lock_set(&lockh, mode, inode);
5371 rc = ll_layout_intent(inode, &intent);
5377 *gen = ll_layout_version_get(lli);
5378 mutex_unlock(&lli->lli_layout_mutex);
5384 * Issue layout intent RPC indicating where in a file an IO is about to write.
5386 * \param[in] inode file inode.
5387 * \param[in] ext write range with start offset of fille in bytes where
5388 * an IO is about to write, and exclusive end offset in
5391 * \retval 0 on success
5392 * \retval < 0 error code
5394 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
5395 struct lu_extent *ext)
5397 struct layout_intent intent = {
5399 .li_extent.e_start = ext->e_start,
5400 .li_extent.e_end = ext->e_end,
5405 rc = ll_layout_intent(inode, &intent);
5411 * This function send a restore request to the MDT
5413 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
5415 struct hsm_user_request *hur;
5419 len = sizeof(struct hsm_user_request) +
5420 sizeof(struct hsm_user_item);
5421 OBD_ALLOC(hur, len);
5425 hur->hur_request.hr_action = HUA_RESTORE;
5426 hur->hur_request.hr_archive_id = 0;
5427 hur->hur_request.hr_flags = 0;
5428 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
5429 sizeof(hur->hur_user_item[0].hui_fid));
5430 hur->hur_user_item[0].hui_extent.offset = offset;
5431 hur->hur_user_item[0].hui_extent.length = length;
5432 hur->hur_request.hr_itemcount = 1;
5433 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,