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 #include <linux/uidgid.h>
47 #include <uapi/linux/lustre/lustre_ioctl.h>
48 #include <lustre_swab.h>
50 #include "cl_object.h"
51 #include "llite_internal.h"
52 #include "vvp_internal.h"
55 struct inode *sp_inode;
60 __u64 pa_data_version;
66 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
68 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
71 static struct ll_file_data *ll_file_data_get(void)
73 struct ll_file_data *fd;
75 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
79 fd->fd_write_failed = false;
80 pcc_file_init(&fd->fd_pcc_file);
85 static void ll_file_data_put(struct ll_file_data *fd)
88 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
92 * Packs all the attributes into @op_data for the CLOSE rpc.
94 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
95 struct obd_client_handle *och)
99 ll_prep_md_op_data(op_data, inode, NULL, NULL,
100 0, 0, LUSTRE_OPC_ANY, NULL);
102 op_data->op_attr.ia_mode = inode->i_mode;
103 op_data->op_attr.ia_atime = inode->i_atime;
104 op_data->op_attr.ia_mtime = inode->i_mtime;
105 op_data->op_attr.ia_ctime = inode->i_ctime;
106 op_data->op_attr.ia_size = i_size_read(inode);
107 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
108 ATTR_MTIME | ATTR_MTIME_SET |
110 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
111 op_data->op_attr_blocks = inode->i_blocks;
112 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
113 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
114 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
115 op_data->op_open_handle = och->och_open_handle;
117 if (och->och_flags & FMODE_WRITE &&
118 ll_file_test_and_clear_flag(ll_i2info(inode), LLIF_DATA_MODIFIED))
119 /* For HSM: if inode data has been modified, pack it so that
120 * MDT can set data dirty flag in the archive. */
121 op_data->op_bias |= MDS_DATA_MODIFIED;
127 * Perform a close, possibly with a bias.
128 * The meaning of "data" depends on the value of "bias".
130 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
131 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
134 static int ll_close_inode_openhandle(struct inode *inode,
135 struct obd_client_handle *och,
136 enum mds_op_bias bias, void *data)
138 struct obd_export *md_exp = ll_i2mdexp(inode);
139 const struct ll_inode_info *lli = ll_i2info(inode);
140 struct md_op_data *op_data;
141 struct ptlrpc_request *req = NULL;
145 if (class_exp2obd(md_exp) == NULL) {
146 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
147 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
151 OBD_ALLOC_PTR(op_data);
152 /* We leak openhandle and request here on error, but not much to be
153 * done in OOM case since app won't retry close on error either. */
155 GOTO(out, rc = -ENOMEM);
157 ll_prepare_close(inode, op_data, och);
159 case MDS_CLOSE_LAYOUT_MERGE:
160 /* merge blocks from the victim inode */
161 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
162 op_data->op_attr.ia_valid |= ATTR_SIZE;
163 op_data->op_xvalid |= OP_XVALID_BLOCKS;
165 case MDS_CLOSE_LAYOUT_SPLIT:
166 case MDS_CLOSE_LAYOUT_SWAP: {
167 struct split_param *sp = data;
169 LASSERT(data != NULL);
170 op_data->op_bias |= bias;
171 op_data->op_data_version = 0;
172 op_data->op_lease_handle = och->och_lease_handle;
173 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
174 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
175 op_data->op_mirror_id = sp->sp_mirror_id;
177 op_data->op_fid2 = *ll_inode2fid(data);
182 case MDS_CLOSE_RESYNC_DONE: {
183 struct ll_ioc_lease *ioc = data;
185 LASSERT(data != NULL);
186 op_data->op_attr_blocks +=
187 ioc->lil_count * op_data->op_attr_blocks;
188 op_data->op_attr.ia_valid |= ATTR_SIZE;
189 op_data->op_xvalid |= OP_XVALID_BLOCKS;
190 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
192 op_data->op_lease_handle = och->och_lease_handle;
193 op_data->op_data = &ioc->lil_ids[0];
194 op_data->op_data_size =
195 ioc->lil_count * sizeof(ioc->lil_ids[0]);
199 case MDS_PCC_ATTACH: {
200 struct pcc_param *param = data;
202 LASSERT(data != NULL);
203 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
204 op_data->op_archive_id = param->pa_archive_id;
205 op_data->op_data_version = param->pa_data_version;
206 op_data->op_lease_handle = och->och_lease_handle;
210 case MDS_HSM_RELEASE:
211 LASSERT(data != NULL);
212 op_data->op_bias |= MDS_HSM_RELEASE;
213 op_data->op_data_version = *(__u64 *)data;
214 op_data->op_lease_handle = och->och_lease_handle;
215 op_data->op_attr.ia_valid |= ATTR_SIZE;
216 op_data->op_xvalid |= OP_XVALID_BLOCKS;
220 LASSERT(data == NULL);
224 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
225 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
226 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
227 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
229 rc = md_close(md_exp, op_data, och->och_mod, &req);
230 if (rc != 0 && rc != -EINTR)
231 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
232 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
234 if (rc == 0 && op_data->op_bias & bias) {
235 struct mdt_body *body;
237 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
238 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
241 if (bias & MDS_PCC_ATTACH) {
242 struct pcc_param *param = data;
244 param->pa_layout_gen = body->mbo_layout_gen;
248 ll_finish_md_op_data(op_data);
252 md_clear_open_replay_data(md_exp, och);
253 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
256 ptlrpc_req_finished(req); /* This is close request */
260 int ll_md_real_close(struct inode *inode, fmode_t fmode)
262 struct ll_inode_info *lli = ll_i2info(inode);
263 struct obd_client_handle **och_p;
264 struct obd_client_handle *och;
269 if (fmode & FMODE_WRITE) {
270 och_p = &lli->lli_mds_write_och;
271 och_usecount = &lli->lli_open_fd_write_count;
272 } else if (fmode & FMODE_EXEC) {
273 och_p = &lli->lli_mds_exec_och;
274 och_usecount = &lli->lli_open_fd_exec_count;
276 LASSERT(fmode & FMODE_READ);
277 och_p = &lli->lli_mds_read_och;
278 och_usecount = &lli->lli_open_fd_read_count;
281 mutex_lock(&lli->lli_och_mutex);
282 if (*och_usecount > 0) {
283 /* There are still users of this handle, so skip
285 mutex_unlock(&lli->lli_och_mutex);
291 mutex_unlock(&lli->lli_och_mutex);
294 /* There might be a race and this handle may already
296 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
302 static int ll_md_close(struct inode *inode, struct file *file)
304 union ldlm_policy_data policy = {
305 .l_inodebits = { MDS_INODELOCK_OPEN },
307 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
308 struct ll_file_data *fd = file->private_data;
309 struct ll_inode_info *lli = ll_i2info(inode);
310 struct lustre_handle lockh;
311 enum ldlm_mode lockmode;
315 /* clear group lock, if present */
316 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
317 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
319 if (fd->fd_lease_och != NULL) {
322 /* Usually the lease is not released when the
323 * application crashed, we need to release here. */
324 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
325 CDEBUG(rc ? D_ERROR : D_INODE, "Clean up lease "DFID" %d/%d\n",
326 PFID(&lli->lli_fid), rc, lease_broken);
328 fd->fd_lease_och = NULL;
331 if (fd->fd_och != NULL) {
332 rc = ll_close_inode_openhandle(inode, fd->fd_och, 0, NULL);
337 /* Let's see if we have good enough OPEN lock on the file and if
338 we can skip talking to MDS */
339 mutex_lock(&lli->lli_och_mutex);
340 if (fd->fd_omode & FMODE_WRITE) {
342 LASSERT(lli->lli_open_fd_write_count);
343 lli->lli_open_fd_write_count--;
344 } else if (fd->fd_omode & FMODE_EXEC) {
346 LASSERT(lli->lli_open_fd_exec_count);
347 lli->lli_open_fd_exec_count--;
350 LASSERT(lli->lli_open_fd_read_count);
351 lli->lli_open_fd_read_count--;
353 mutex_unlock(&lli->lli_och_mutex);
355 /* LU-4398: do not cache write open lock if the file has exec bit */
356 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
357 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
358 LDLM_IBITS, &policy, lockmode, &lockh))
359 rc = ll_md_real_close(inode, fd->fd_omode);
362 file->private_data = NULL;
363 ll_file_data_put(fd);
368 /* While this returns an error code, fput() the caller does not, so we need
369 * to make every effort to clean up all of our state here. Also, applications
370 * rarely check close errors and even if an error is returned they will not
371 * re-try the close call.
373 int ll_file_release(struct inode *inode, struct file *file)
375 struct ll_file_data *fd;
376 struct ll_sb_info *sbi = ll_i2sbi(inode);
377 struct ll_inode_info *lli = ll_i2info(inode);
378 ktime_t kstart = ktime_get();
383 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
384 PFID(ll_inode2fid(inode)), inode);
386 fd = file->private_data;
389 /* The last ref on @file, maybe not the the owner pid of statahead,
390 * because parent and child process can share the same file handle. */
391 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
392 ll_deauthorize_statahead(inode, fd);
394 if (inode->i_sb->s_root == file_dentry(file)) {
395 file->private_data = NULL;
396 ll_file_data_put(fd);
400 pcc_file_release(inode, file);
402 if (!S_ISDIR(inode->i_mode)) {
403 if (lli->lli_clob != NULL)
404 lov_read_and_clear_async_rc(lli->lli_clob);
405 lli->lli_async_rc = 0;
408 rc = ll_md_close(inode, file);
410 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
411 libcfs_debug_dumplog();
414 if (!rc && inode->i_sb->s_root != file_dentry(file))
415 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
416 ktime_us_delta(ktime_get(), kstart));
420 static inline int ll_dom_readpage(void *data, struct page *page)
422 struct niobuf_local *lnb = data;
425 kaddr = kmap_atomic(page);
426 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
427 if (lnb->lnb_len < PAGE_SIZE)
428 memset(kaddr + lnb->lnb_len, 0,
429 PAGE_SIZE - lnb->lnb_len);
430 flush_dcache_page(page);
431 SetPageUptodate(page);
432 kunmap_atomic(kaddr);
438 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req,
439 struct lookup_intent *it)
441 struct ll_inode_info *lli = ll_i2info(inode);
442 struct cl_object *obj = lli->lli_clob;
443 struct address_space *mapping = inode->i_mapping;
445 struct niobuf_remote *rnb;
446 struct mdt_body *body;
448 unsigned long index, start;
449 struct niobuf_local lnb;
456 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
460 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
461 if (rnb == NULL || rnb->rnb_len == 0)
464 /* LU-11595: Server may return whole file and that is OK always or
465 * it may return just file tail and its offset must be aligned with
466 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
467 * smaller then offset may be not aligned and that data is just ignored.
469 if (rnb->rnb_offset & ~PAGE_MASK)
472 /* Server returns whole file or just file tail if it fills in reply
473 * buffer, in both cases total size should be equal to the file size.
475 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
476 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size) {
477 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
478 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
479 rnb->rnb_len, body->mbo_dom_size);
483 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
484 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
486 data = (char *)rnb + sizeof(*rnb);
488 lnb.lnb_file_offset = rnb->rnb_offset;
489 start = lnb.lnb_file_offset >> PAGE_SHIFT;
491 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
492 lnb.lnb_page_offset = 0;
494 lnb.lnb_data = data + (index << PAGE_SHIFT);
495 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
496 if (lnb.lnb_len > PAGE_SIZE)
497 lnb.lnb_len = PAGE_SIZE;
499 vmpage = read_cache_page(mapping, index + start,
500 ll_dom_readpage, &lnb);
501 if (IS_ERR(vmpage)) {
502 CWARN("%s: cannot fill page %lu for "DFID
503 " with data: rc = %li\n",
504 ll_i2sbi(inode)->ll_fsname, index + start,
505 PFID(lu_object_fid(&obj->co_lu)),
511 } while (rnb->rnb_len > (index << PAGE_SHIFT));
515 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
516 struct lookup_intent *itp)
518 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
519 struct dentry *parent = de->d_parent;
522 struct md_op_data *op_data;
523 struct ptlrpc_request *req = NULL;
527 LASSERT(parent != NULL);
528 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
530 /* if server supports open-by-fid, or file name is invalid, don't pack
531 * name in open request */
532 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
533 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
535 len = de->d_name.len;
536 name = kmalloc(len + 1, GFP_NOFS);
541 spin_lock(&de->d_lock);
542 if (len != de->d_name.len) {
543 spin_unlock(&de->d_lock);
547 memcpy(name, de->d_name.name, len);
549 spin_unlock(&de->d_lock);
551 if (!lu_name_is_valid_2(name, len)) {
557 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
558 name, len, 0, LUSTRE_OPC_ANY, NULL);
559 if (IS_ERR(op_data)) {
561 RETURN(PTR_ERR(op_data));
563 op_data->op_data = lmm;
564 op_data->op_data_size = lmmsize;
566 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
567 &ll_md_blocking_ast, 0);
569 ll_finish_md_op_data(op_data);
571 /* reason for keep own exit path - don`t flood log
572 * with messages with -ESTALE errors.
574 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
575 it_open_error(DISP_OPEN_OPEN, itp))
577 ll_release_openhandle(de, itp);
581 if (it_disposition(itp, DISP_LOOKUP_NEG))
582 GOTO(out, rc = -ENOENT);
584 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
585 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
586 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
590 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
592 if (!rc && itp->it_lock_mode) {
593 struct lustre_handle handle = {.cookie = itp->it_lock_handle};
594 struct ldlm_lock *lock;
595 bool has_dom_bit = false;
597 /* If we got a lock back and it has a LOOKUP bit set,
598 * make sure the dentry is marked as valid so we can find it.
599 * We don't need to care about actual hashing since other bits
600 * of kernel will deal with that later.
602 lock = ldlm_handle2lock(&handle);
604 has_dom_bit = ldlm_has_dom(lock);
605 if (lock->l_policy_data.l_inodebits.bits &
606 MDS_INODELOCK_LOOKUP)
607 d_lustre_revalidate(de);
611 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, NULL);
613 ll_dom_finish_open(de->d_inode, req, itp);
617 ptlrpc_req_finished(req);
618 ll_intent_drop_lock(itp);
620 /* We did open by fid, but by the time we got to the server,
621 * the object disappeared. If this is a create, we cannot really
622 * tell the userspace that the file it was trying to create
623 * does not exist. Instead let's return -ESTALE, and the VFS will
624 * retry the create with LOOKUP_REVAL that we are going to catch
625 * in ll_revalidate_dentry() and use lookup then.
627 if (rc == -ENOENT && itp->it_op & IT_CREAT)
633 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
634 struct obd_client_handle *och)
636 struct mdt_body *body;
638 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
639 och->och_open_handle = body->mbo_open_handle;
640 och->och_fid = body->mbo_fid1;
641 och->och_lease_handle.cookie = it->it_lock_handle;
642 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
643 och->och_flags = it->it_flags;
645 return md_set_open_replay_data(md_exp, och, it);
648 static int ll_local_open(struct file *file, struct lookup_intent *it,
649 struct ll_file_data *fd, struct obd_client_handle *och)
651 struct inode *inode = file_inode(file);
654 LASSERT(!file->private_data);
661 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
666 file->private_data = fd;
667 ll_readahead_init(inode, &fd->fd_ras);
668 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
670 /* ll_cl_context initialize */
671 rwlock_init(&fd->fd_lock);
672 INIT_LIST_HEAD(&fd->fd_lccs);
677 /* Open a file, and (for the very first open) create objects on the OSTs at
678 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
679 * creation or open until ll_lov_setstripe() ioctl is called.
681 * If we already have the stripe MD locally then we don't request it in
682 * md_open(), by passing a lmm_size = 0.
684 * It is up to the application to ensure no other processes open this file
685 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
686 * used. We might be able to avoid races of that sort by getting lli_open_sem
687 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
688 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
690 int ll_file_open(struct inode *inode, struct file *file)
692 struct ll_inode_info *lli = ll_i2info(inode);
693 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
694 .it_flags = file->f_flags };
695 struct obd_client_handle **och_p = NULL;
696 __u64 *och_usecount = NULL;
697 struct ll_file_data *fd;
698 ktime_t kstart = ktime_get();
702 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
703 PFID(ll_inode2fid(inode)), inode, file->f_flags);
705 it = file->private_data; /* XXX: compat macro */
706 file->private_data = NULL; /* prevent ll_local_open assertion */
708 fd = ll_file_data_get();
710 GOTO(out_nofiledata, rc = -ENOMEM);
713 if (S_ISDIR(inode->i_mode))
714 ll_authorize_statahead(inode, fd);
716 if (inode->i_sb->s_root == file_dentry(file)) {
717 file->private_data = fd;
721 if (!it || !it->it_disposition) {
722 /* Convert f_flags into access mode. We cannot use file->f_mode,
723 * because everything but O_ACCMODE mask was stripped from
725 if ((oit.it_flags + 1) & O_ACCMODE)
727 if (file->f_flags & O_TRUNC)
728 oit.it_flags |= FMODE_WRITE;
730 /* kernel only call f_op->open in dentry_open. filp_open calls
731 * dentry_open after call to open_namei that checks permissions.
732 * Only nfsd_open call dentry_open directly without checking
733 * permissions and because of that this code below is safe.
735 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
736 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
738 /* We do not want O_EXCL here, presumably we opened the file
739 * already? XXX - NFS implications? */
740 oit.it_flags &= ~O_EXCL;
742 /* bug20584, if "it_flags" contains O_CREAT, the file will be
743 * created if necessary, then "IT_CREAT" should be set to keep
744 * consistent with it */
745 if (oit.it_flags & O_CREAT)
746 oit.it_op |= IT_CREAT;
752 /* Let's see if we have file open on MDS already. */
753 if (it->it_flags & FMODE_WRITE) {
754 och_p = &lli->lli_mds_write_och;
755 och_usecount = &lli->lli_open_fd_write_count;
756 } else if (it->it_flags & FMODE_EXEC) {
757 och_p = &lli->lli_mds_exec_och;
758 och_usecount = &lli->lli_open_fd_exec_count;
760 och_p = &lli->lli_mds_read_och;
761 och_usecount = &lli->lli_open_fd_read_count;
764 mutex_lock(&lli->lli_och_mutex);
765 if (*och_p) { /* Open handle is present */
766 if (it_disposition(it, DISP_OPEN_OPEN)) {
767 /* Well, there's extra open request that we do not need,
768 * let's close it somehow. This will decref request. */
769 rc = it_open_error(DISP_OPEN_OPEN, it);
771 mutex_unlock(&lli->lli_och_mutex);
772 GOTO(out_openerr, rc);
775 ll_release_openhandle(file_dentry(file), it);
779 rc = ll_local_open(file, it, fd, NULL);
782 mutex_unlock(&lli->lli_och_mutex);
783 GOTO(out_openerr, rc);
786 LASSERT(*och_usecount == 0);
787 if (!it->it_disposition) {
788 struct dentry *dentry = file_dentry(file);
789 struct ll_dentry_data *ldd;
791 /* We cannot just request lock handle now, new ELC code
792 * means that one of other OPEN locks for this file
793 * could be cancelled, and since blocking ast handler
794 * would attempt to grab och_mutex as well, that would
795 * result in a deadlock
797 mutex_unlock(&lli->lli_och_mutex);
799 * Normally called under two situations:
801 * 2. A race/condition on MDS resulting in no open
802 * handle to be returned from LOOKUP|OPEN request,
803 * for example if the target entry was a symlink.
805 * Only fetch MDS_OPEN_LOCK if this is in NFS path,
806 * marked by a bit set in ll_iget_for_nfs. Clear the
807 * bit so that it's not confusing later callers.
809 * NB; when ldd is NULL, it must have come via normal
810 * lookup path only, since ll_iget_for_nfs always calls
813 ldd = ll_d2d(dentry);
814 if (ldd && ldd->lld_nfs_dentry) {
815 ldd->lld_nfs_dentry = 0;
816 if (!filename_is_volatile(dentry->d_name.name,
819 it->it_flags |= MDS_OPEN_LOCK;
823 * Always specify MDS_OPEN_BY_FID because we don't want
824 * to get file with different fid.
826 it->it_flags |= MDS_OPEN_BY_FID;
827 rc = ll_intent_file_open(dentry, NULL, 0, it);
829 GOTO(out_openerr, rc);
833 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
835 GOTO(out_och_free, rc = -ENOMEM);
839 /* md_intent_lock() didn't get a request ref if there was an
840 * open error, so don't do cleanup on the request here
842 /* XXX (green): Should not we bail out on any error here, not
843 * just open error? */
844 rc = it_open_error(DISP_OPEN_OPEN, it);
846 GOTO(out_och_free, rc);
848 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
849 "inode %p: disposition %x, status %d\n", inode,
850 it_disposition(it, ~0), it->it_status);
852 rc = ll_local_open(file, it, fd, *och_p);
854 GOTO(out_och_free, rc);
857 rc = pcc_file_open(inode, file);
859 GOTO(out_och_free, rc);
861 mutex_unlock(&lli->lli_och_mutex);
863 /* lockless for direct IO so that it can do IO in parallel */
864 if (file->f_flags & O_DIRECT)
865 fd->fd_flags |= LL_FILE_LOCKLESS_IO;
868 /* Must do this outside lli_och_mutex lock to prevent deadlock where
869 different kind of OPEN lock for this same inode gets cancelled
870 by ldlm_cancel_lru */
871 if (!S_ISREG(inode->i_mode))
872 GOTO(out_och_free, rc);
873 cl_lov_delay_create_clear(&file->f_flags);
874 GOTO(out_och_free, rc);
878 if (och_p && *och_p) {
879 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
880 *och_p = NULL; /* OBD_FREE writes some magic there */
883 mutex_unlock(&lli->lli_och_mutex);
886 if (lli->lli_opendir_key == fd)
887 ll_deauthorize_statahead(inode, fd);
890 ll_file_data_put(fd);
892 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
893 ktime_us_delta(ktime_get(), kstart));
897 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
898 ptlrpc_req_finished(it->it_request);
899 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
905 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
906 struct ldlm_lock_desc *desc, void *data, int flag)
909 struct lustre_handle lockh;
913 case LDLM_CB_BLOCKING:
914 ldlm_lock2handle(lock, &lockh);
915 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
917 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
921 case LDLM_CB_CANCELING:
929 * When setting a lease on a file, we take ownership of the lli_mds_*_och
930 * and save it as fd->fd_och so as to force client to reopen the file even
931 * if it has an open lock in cache already.
933 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
934 struct lustre_handle *old_open_handle)
936 struct ll_inode_info *lli = ll_i2info(inode);
937 struct ll_file_data *fd = file->private_data;
938 struct obd_client_handle **och_p;
943 /* Get the openhandle of the file */
944 mutex_lock(&lli->lli_och_mutex);
945 if (fd->fd_lease_och != NULL)
946 GOTO(out_unlock, rc = -EBUSY);
948 if (fd->fd_och == NULL) {
949 if (file->f_mode & FMODE_WRITE) {
950 LASSERT(lli->lli_mds_write_och != NULL);
951 och_p = &lli->lli_mds_write_och;
952 och_usecount = &lli->lli_open_fd_write_count;
954 LASSERT(lli->lli_mds_read_och != NULL);
955 och_p = &lli->lli_mds_read_och;
956 och_usecount = &lli->lli_open_fd_read_count;
959 if (*och_usecount > 1)
960 GOTO(out_unlock, rc = -EBUSY);
967 *old_open_handle = fd->fd_och->och_open_handle;
971 mutex_unlock(&lli->lli_och_mutex);
976 * Release ownership on lli_mds_*_och when putting back a file lease.
978 static int ll_lease_och_release(struct inode *inode, struct file *file)
980 struct ll_inode_info *lli = ll_i2info(inode);
981 struct ll_file_data *fd = file->private_data;
982 struct obd_client_handle **och_p;
983 struct obd_client_handle *old_och = NULL;
988 mutex_lock(&lli->lli_och_mutex);
989 if (file->f_mode & FMODE_WRITE) {
990 och_p = &lli->lli_mds_write_och;
991 och_usecount = &lli->lli_open_fd_write_count;
993 och_p = &lli->lli_mds_read_och;
994 och_usecount = &lli->lli_open_fd_read_count;
997 /* The file may have been open by another process (broken lease) so
998 * *och_p is not NULL. In this case we should simply increase usecount
1001 if (*och_p != NULL) {
1002 old_och = fd->fd_och;
1005 *och_p = fd->fd_och;
1009 mutex_unlock(&lli->lli_och_mutex);
1011 if (old_och != NULL)
1012 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1018 * Acquire a lease and open the file.
1020 static struct obd_client_handle *
1021 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1024 struct lookup_intent it = { .it_op = IT_OPEN };
1025 struct ll_sb_info *sbi = ll_i2sbi(inode);
1026 struct md_op_data *op_data;
1027 struct ptlrpc_request *req = NULL;
1028 struct lustre_handle old_open_handle = { 0 };
1029 struct obd_client_handle *och = NULL;
1034 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1035 RETURN(ERR_PTR(-EINVAL));
1038 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1039 RETURN(ERR_PTR(-EPERM));
1041 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1043 RETURN(ERR_PTR(rc));
1048 RETURN(ERR_PTR(-ENOMEM));
1050 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1051 LUSTRE_OPC_ANY, NULL);
1052 if (IS_ERR(op_data))
1053 GOTO(out, rc = PTR_ERR(op_data));
1055 /* To tell the MDT this openhandle is from the same owner */
1056 op_data->op_open_handle = old_open_handle;
1058 it.it_flags = fmode | open_flags;
1059 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1060 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1061 &ll_md_blocking_lease_ast,
1062 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1063 * it can be cancelled which may mislead applications that the lease is
1065 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1066 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1067 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1068 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1069 ll_finish_md_op_data(op_data);
1070 ptlrpc_req_finished(req);
1072 GOTO(out_release_it, rc);
1074 if (it_disposition(&it, DISP_LOOKUP_NEG))
1075 GOTO(out_release_it, rc = -ENOENT);
1077 rc = it_open_error(DISP_OPEN_OPEN, &it);
1079 GOTO(out_release_it, rc);
1081 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1082 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1084 GOTO(out_release_it, rc);
1086 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1087 GOTO(out_close, rc = -EOPNOTSUPP);
1089 /* already get lease, handle lease lock */
1090 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1091 if (it.it_lock_mode == 0 ||
1092 it.it_lock_bits != MDS_INODELOCK_OPEN) {
1093 /* open lock must return for lease */
1094 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1095 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1097 GOTO(out_close, rc = -EPROTO);
1100 ll_intent_release(&it);
1104 /* Cancel open lock */
1105 if (it.it_lock_mode != 0) {
1106 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1108 it.it_lock_mode = 0;
1109 och->och_lease_handle.cookie = 0ULL;
1111 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1113 CERROR("%s: error closing file "DFID": %d\n",
1114 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1115 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1117 ll_intent_release(&it);
1121 RETURN(ERR_PTR(rc));
1125 * Check whether a layout swap can be done between two inodes.
1127 * \param[in] inode1 First inode to check
1128 * \param[in] inode2 Second inode to check
1130 * \retval 0 on success, layout swap can be performed between both inodes
1131 * \retval negative error code if requirements are not met
1133 static int ll_check_swap_layouts_validity(struct inode *inode1,
1134 struct inode *inode2)
1136 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1139 if (inode_permission(inode1, MAY_WRITE) ||
1140 inode_permission(inode2, MAY_WRITE))
1143 if (inode1->i_sb != inode2->i_sb)
1149 static int ll_swap_layouts_close(struct obd_client_handle *och,
1150 struct inode *inode, struct inode *inode2)
1152 const struct lu_fid *fid1 = ll_inode2fid(inode);
1153 const struct lu_fid *fid2;
1157 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1158 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1160 rc = ll_check_swap_layouts_validity(inode, inode2);
1162 GOTO(out_free_och, rc);
1164 /* We now know that inode2 is a lustre inode */
1165 fid2 = ll_inode2fid(inode2);
1167 rc = lu_fid_cmp(fid1, fid2);
1169 GOTO(out_free_och, rc = -EINVAL);
1171 /* Close the file and {swap,merge} layouts between inode & inode2.
1172 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1173 * because we still need it to pack l_remote_handle to MDT. */
1174 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1177 och = NULL; /* freed in ll_close_inode_openhandle() */
1187 * Release lease and close the file.
1188 * It will check if the lease has ever broken.
1190 static int ll_lease_close_intent(struct obd_client_handle *och,
1191 struct inode *inode,
1192 bool *lease_broken, enum mds_op_bias bias,
1195 struct ldlm_lock *lock;
1196 bool cancelled = true;
1200 lock = ldlm_handle2lock(&och->och_lease_handle);
1202 lock_res_and_lock(lock);
1203 cancelled = ldlm_is_cancel(lock);
1204 unlock_res_and_lock(lock);
1205 LDLM_LOCK_PUT(lock);
1208 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1209 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1211 if (lease_broken != NULL)
1212 *lease_broken = cancelled;
1214 if (!cancelled && !bias)
1215 ldlm_cli_cancel(&och->och_lease_handle, 0);
1217 if (cancelled) { /* no need to excute intent */
1222 rc = ll_close_inode_openhandle(inode, och, bias, data);
1226 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1229 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1233 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1235 static int ll_lease_file_resync(struct obd_client_handle *och,
1236 struct inode *inode, unsigned long arg)
1238 struct ll_sb_info *sbi = ll_i2sbi(inode);
1239 struct md_op_data *op_data;
1240 struct ll_ioc_lease_id ioc;
1241 __u64 data_version_unused;
1245 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1246 LUSTRE_OPC_ANY, NULL);
1247 if (IS_ERR(op_data))
1248 RETURN(PTR_ERR(op_data));
1250 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1254 /* before starting file resync, it's necessary to clean up page cache
1255 * in client memory, otherwise once the layout version is increased,
1256 * writing back cached data will be denied the OSTs. */
1257 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1261 op_data->op_lease_handle = och->och_lease_handle;
1262 op_data->op_mirror_id = ioc.lil_mirror_id;
1263 rc = md_file_resync(sbi->ll_md_exp, op_data);
1269 ll_finish_md_op_data(op_data);
1273 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1275 struct ll_inode_info *lli = ll_i2info(inode);
1276 struct cl_object *obj = lli->lli_clob;
1277 struct cl_attr *attr = vvp_env_thread_attr(env);
1285 ll_inode_size_lock(inode);
1287 /* Merge timestamps the most recently obtained from MDS with
1288 * timestamps obtained from OSTs.
1290 * Do not overwrite atime of inode because it may be refreshed
1291 * by file_accessed() function. If the read was served by cache
1292 * data, there is no RPC to be sent so that atime may not be
1293 * transferred to OSTs at all. MDT only updates atime at close time
1294 * if it's at least 'mdd.*.atime_diff' older.
1295 * All in all, the atime in Lustre does not strictly comply with
1296 * POSIX. Solving this problem needs to send an RPC to MDT for each
1297 * read, this will hurt performance.
1299 if (ll_file_test_and_clear_flag(lli, LLIF_UPDATE_ATIME) ||
1300 inode->i_atime.tv_sec < lli->lli_atime)
1301 inode->i_atime.tv_sec = lli->lli_atime;
1303 inode->i_mtime.tv_sec = lli->lli_mtime;
1304 inode->i_ctime.tv_sec = lli->lli_ctime;
1306 mtime = inode->i_mtime.tv_sec;
1307 atime = inode->i_atime.tv_sec;
1308 ctime = inode->i_ctime.tv_sec;
1310 cl_object_attr_lock(obj);
1311 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1314 rc = cl_object_attr_get(env, obj, attr);
1315 cl_object_attr_unlock(obj);
1318 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1320 if (atime < attr->cat_atime)
1321 atime = attr->cat_atime;
1323 if (ctime < attr->cat_ctime)
1324 ctime = attr->cat_ctime;
1326 if (mtime < attr->cat_mtime)
1327 mtime = attr->cat_mtime;
1329 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1330 PFID(&lli->lli_fid), attr->cat_size);
1332 i_size_write(inode, attr->cat_size);
1333 inode->i_blocks = attr->cat_blocks;
1335 inode->i_mtime.tv_sec = mtime;
1336 inode->i_atime.tv_sec = atime;
1337 inode->i_ctime.tv_sec = ctime;
1340 ll_inode_size_unlock(inode);
1346 * Set designated mirror for I/O.
1348 * So far only read, write, and truncated can support to issue I/O to
1349 * designated mirror.
1351 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1353 struct ll_file_data *fd = file->private_data;
1355 /* clear layout version for generic(non-resync) I/O in case it carries
1356 * stale layout version due to I/O restart */
1357 io->ci_layout_version = 0;
1359 /* FLR: disable non-delay for designated mirror I/O because obviously
1360 * only one mirror is available */
1361 if (fd->fd_designated_mirror > 0) {
1363 io->ci_designated_mirror = fd->fd_designated_mirror;
1364 io->ci_layout_version = fd->fd_layout_version;
1367 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1368 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1371 static bool file_is_noatime(const struct file *file)
1373 const struct vfsmount *mnt = file->f_path.mnt;
1374 const struct inode *inode = file_inode((struct file *)file);
1376 /* Adapted from file_accessed() and touch_atime().*/
1377 if (file->f_flags & O_NOATIME)
1380 if (inode->i_flags & S_NOATIME)
1383 if (IS_NOATIME(inode))
1386 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1389 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1392 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1398 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1399 struct vvp_io_args *args)
1401 struct inode *inode = file_inode(file);
1402 struct ll_file_data *fd = file->private_data;
1404 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1405 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1407 if (iot == CIT_WRITE) {
1408 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1409 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1410 file->f_flags & O_DIRECT ||
1412 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1413 io->u.ci_wr.wr_sync |= !!(args &&
1414 args->via_io_subtype == IO_NORMAL &&
1415 args->u.normal.via_iocb->ki_flags & IOCB_DSYNC);
1419 io->ci_obj = ll_i2info(inode)->lli_clob;
1420 io->ci_lockreq = CILR_MAYBE;
1421 if (ll_file_nolock(file)) {
1422 io->ci_lockreq = CILR_NEVER;
1423 io->ci_no_srvlock = 1;
1424 } else if (file->f_flags & O_APPEND) {
1425 io->ci_lockreq = CILR_MANDATORY;
1427 io->ci_noatime = file_is_noatime(file);
1428 io->ci_async_readahead = false;
1430 /* FLR: only use non-delay I/O for read as there is only one
1431 * avaliable mirror for write. */
1432 io->ci_ndelay = !(iot == CIT_WRITE);
1434 ll_io_set_mirror(io, file);
1437 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1440 struct ll_inode_info *lli = ll_i2info(inode);
1441 struct ll_sb_info *sbi = ll_i2sbi(inode);
1442 enum obd_heat_type sample_type;
1443 enum obd_heat_type iobyte_type;
1444 __u64 now = ktime_get_real_seconds();
1446 if (!ll_sbi_has_file_heat(sbi) ||
1447 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1450 if (iot == CIT_READ) {
1451 sample_type = OBD_HEAT_READSAMPLE;
1452 iobyte_type = OBD_HEAT_READBYTE;
1453 } else if (iot == CIT_WRITE) {
1454 sample_type = OBD_HEAT_WRITESAMPLE;
1455 iobyte_type = OBD_HEAT_WRITEBYTE;
1460 spin_lock(&lli->lli_heat_lock);
1461 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1462 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1463 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1464 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1465 spin_unlock(&lli->lli_heat_lock);
1469 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1470 struct file *file, enum cl_io_type iot,
1471 loff_t *ppos, size_t count)
1473 struct vvp_io *vio = vvp_env_io(env);
1474 struct inode *inode = file_inode(file);
1475 struct ll_inode_info *lli = ll_i2info(inode);
1476 struct ll_file_data *fd = file->private_data;
1477 struct range_lock range;
1481 unsigned retried = 0;
1482 unsigned ignore_lockless = 0;
1486 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1487 file_dentry(file)->d_name.name,
1488 iot == CIT_READ ? "read" : "write", *ppos, count);
1491 io = vvp_env_thread_io(env);
1492 ll_io_init(io, file, iot, args);
1493 io->ci_ignore_lockless = ignore_lockless;
1494 io->ci_ndelay_tried = retried;
1496 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1497 bool range_locked = false;
1499 if (file->f_flags & O_APPEND)
1500 range_lock_init(&range, 0, LUSTRE_EOF);
1502 range_lock_init(&range, *ppos, *ppos + count - 1);
1504 vio->vui_fd = file->private_data;
1505 vio->vui_io_subtype = args->via_io_subtype;
1507 switch (vio->vui_io_subtype) {
1509 vio->vui_iter = args->u.normal.via_iter;
1510 vio->vui_iocb = args->u.normal.via_iocb;
1511 /* Direct IO reads must also take range lock,
1512 * or multiple reads will try to work on the same pages
1513 * See LU-6227 for details. */
1514 if (((iot == CIT_WRITE) ||
1515 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1516 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1517 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1519 rc = range_lock(&lli->lli_write_tree, &range);
1523 range_locked = true;
1527 vio->u.splice.vui_pipe = args->u.splice.via_pipe;
1528 vio->u.splice.vui_flags = args->u.splice.via_flags;
1531 CERROR("unknown IO subtype %u\n", vio->vui_io_subtype);
1535 ll_cl_add(file, env, io, LCC_RW);
1536 rc = cl_io_loop(env, io);
1537 ll_cl_remove(file, env);
1540 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1542 range_unlock(&lli->lli_write_tree, &range);
1545 /* cl_io_rw_init() handled IO */
1549 if (io->ci_nob > 0) {
1550 result += io->ci_nob;
1551 count -= io->ci_nob;
1552 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1554 /* prepare IO restart */
1555 if (count > 0 && args->via_io_subtype == IO_NORMAL)
1556 args->u.normal.via_iter = vio->vui_iter;
1559 cl_io_fini(env, io);
1562 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1563 file->f_path.dentry->d_name.name,
1564 iot, rc, result, io->ci_need_restart);
1566 if ((rc == 0 || rc == -ENODATA || rc == -ENOLCK) &&
1567 count > 0 && io->ci_need_restart) {
1569 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1570 file_dentry(file)->d_name.name,
1571 iot == CIT_READ ? "read" : "write",
1572 *ppos, count, result, rc);
1573 /* preserve the tried count for FLR */
1574 retried = io->ci_ndelay_tried;
1575 ignore_lockless = io->ci_ignore_lockless;
1579 if (iot == CIT_READ) {
1581 ll_stats_ops_tally(ll_i2sbi(inode),
1582 LPROC_LL_READ_BYTES, result);
1583 } else if (iot == CIT_WRITE) {
1585 ll_stats_ops_tally(ll_i2sbi(inode),
1586 LPROC_LL_WRITE_BYTES, result);
1587 fd->fd_write_failed = false;
1588 } else if (result == 0 && rc == 0) {
1591 fd->fd_write_failed = true;
1593 fd->fd_write_failed = false;
1594 } else if (rc != -ERESTARTSYS) {
1595 fd->fd_write_failed = true;
1599 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1601 ll_heat_add(inode, iot, result);
1603 RETURN(result > 0 ? result : rc);
1607 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1608 * especially for small I/O.
1610 * To serve a read request, CLIO has to create and initialize a cl_io and
1611 * then request DLM lock. This has turned out to have siginificant overhead
1612 * and affects the performance of small I/O dramatically.
1614 * It's not necessary to create a cl_io for each I/O. Under the help of read
1615 * ahead, most of the pages being read are already in memory cache and we can
1616 * read those pages directly because if the pages exist, the corresponding DLM
1617 * lock must exist so that page content must be valid.
1619 * In fast read implementation, the llite speculatively finds and reads pages
1620 * in memory cache. There are three scenarios for fast read:
1621 * - If the page exists and is uptodate, kernel VM will provide the data and
1622 * CLIO won't be intervened;
1623 * - If the page was brought into memory by read ahead, it will be exported
1624 * and read ahead parameters will be updated;
1625 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1626 * it will go back and invoke normal read, i.e., a cl_io will be created
1627 * and DLM lock will be requested.
1629 * POSIX compliance: posix standard states that read is intended to be atomic.
1630 * Lustre read implementation is in line with Linux kernel read implementation
1631 * and neither of them complies with POSIX standard in this matter. Fast read
1632 * doesn't make the situation worse on single node but it may interleave write
1633 * results from multiple nodes due to short read handling in ll_file_aio_read().
1635 * \param env - lu_env
1636 * \param iocb - kiocb from kernel
1637 * \param iter - user space buffers where the data will be copied
1639 * \retval - number of bytes have been read, or error code if error occurred.
1642 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1646 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1649 /* NB: we can't do direct IO for fast read because it will need a lock
1650 * to make IO engine happy. */
1651 if (iocb->ki_filp->f_flags & O_DIRECT)
1654 result = generic_file_read_iter(iocb, iter);
1656 /* If the first page is not in cache, generic_file_aio_read() will be
1657 * returned with -ENODATA.
1658 * See corresponding code in ll_readpage(). */
1659 if (result == -ENODATA)
1663 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1664 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1665 LPROC_LL_READ_BYTES, result);
1672 * Read from a file (through the page cache).
1674 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1677 struct vvp_io_args *args;
1678 struct file *file = iocb->ki_filp;
1682 ktime_t kstart = ktime_get();
1685 if (!iov_iter_count(to))
1689 * Currently when PCC read failed, we do not fall back to the
1690 * normal read path, just return the error.
1691 * The resaon is that: for RW-PCC, the file data may be modified
1692 * in the PCC and inconsistent with the data on OSTs (or file
1693 * data has been removed from the Lustre file system), at this
1694 * time, fallback to the normal read path may read the wrong
1696 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1697 * path: read data from data copy on OSTs.
1699 result = pcc_file_read_iter(iocb, to, &cached);
1703 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
1705 result = ll_do_fast_read(iocb, to);
1706 if (result < 0 || iov_iter_count(to) == 0)
1709 env = cl_env_get(&refcheck);
1711 return PTR_ERR(env);
1713 args = ll_env_args(env, IO_NORMAL);
1714 args->u.normal.via_iter = to;
1715 args->u.normal.via_iocb = iocb;
1717 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
1718 &iocb->ki_pos, iov_iter_count(to));
1721 else if (result == 0)
1724 cl_env_put(env, &refcheck);
1727 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1728 file->private_data, iocb->ki_pos, result,
1730 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
1731 ktime_us_delta(ktime_get(), kstart));
1738 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
1739 * If a page is already in the page cache and dirty (and some other things -
1740 * See ll_tiny_write_begin for the instantiation of these rules), then we can
1741 * write to it without doing a full I/O, because Lustre already knows about it
1742 * and will write it out. This saves a lot of processing time.
1744 * All writes here are within one page, so exclusion is handled by the page
1745 * lock on the vm page. We do not do tiny writes for writes which touch
1746 * multiple pages because it's very unlikely multiple sequential pages are
1747 * are already dirty.
1749 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
1750 * and are unlikely to be to already dirty pages.
1752 * Attribute updates are important here, we do them in ll_tiny_write_end.
1754 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
1756 ssize_t count = iov_iter_count(iter);
1757 struct file *file = iocb->ki_filp;
1758 struct inode *inode = file_inode(file);
1759 bool lock_inode = !IS_NOSEC(inode);
1764 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
1765 * of function for why.
1767 if (count >= PAGE_SIZE ||
1768 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
1771 if (unlikely(lock_inode))
1773 result = __generic_file_write_iter(iocb, iter);
1775 if (unlikely(lock_inode))
1776 inode_unlock(inode);
1778 /* If the page is not already dirty, ll_tiny_write_begin returns
1779 * -ENODATA. We continue on to normal write.
1781 if (result == -ENODATA)
1785 ll_heat_add(inode, CIT_WRITE, result);
1786 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES,
1788 ll_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED);
1791 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
1797 * Write to a file (through the page cache).
1799 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1801 struct vvp_io_args *args;
1803 ssize_t rc_tiny = 0, rc_normal;
1804 struct file *file = iocb->ki_filp;
1807 ktime_t kstart = ktime_get();
1812 if (!iov_iter_count(from))
1813 GOTO(out, rc_normal = 0);
1816 * When PCC write failed, we usually do not fall back to the normal
1817 * write path, just return the error. But there is a special case when
1818 * returned error code is -ENOSPC due to running out of space on PCC HSM
1819 * bakcend. At this time, it will fall back to normal I/O path and
1820 * retry the I/O. As the file is in HSM released state, it will restore
1821 * the file data to OSTs first and redo the write again. And the
1822 * restore process will revoke the layout lock and detach the file
1823 * from PCC cache automatically.
1825 result = pcc_file_write_iter(iocb, from, &cached);
1826 if (cached && result != -ENOSPC && result != -EDQUOT)
1827 GOTO(out, rc_normal = result);
1829 /* NB: we can't do direct IO for tiny writes because they use the page
1830 * cache, we can't do sync writes because tiny writes can't flush
1831 * pages, and we can't do append writes because we can't guarantee the
1832 * required DLM locks are held to protect file size.
1834 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
1835 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
1836 rc_tiny = ll_do_tiny_write(iocb, from);
1838 /* In case of error, go on and try normal write - Only stop if tiny
1839 * write completed I/O.
1841 if (iov_iter_count(from) == 0)
1842 GOTO(out, rc_normal = rc_tiny);
1844 env = cl_env_get(&refcheck);
1846 return PTR_ERR(env);
1848 args = ll_env_args(env, IO_NORMAL);
1849 args->u.normal.via_iter = from;
1850 args->u.normal.via_iocb = iocb;
1852 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
1853 &iocb->ki_pos, iov_iter_count(from));
1855 /* On success, combine bytes written. */
1856 if (rc_tiny >= 0 && rc_normal > 0)
1857 rc_normal += rc_tiny;
1858 /* On error, only return error from normal write if tiny write did not
1859 * write any bytes. Otherwise return bytes written by tiny write.
1861 else if (rc_tiny > 0)
1862 rc_normal = rc_tiny;
1864 cl_env_put(env, &refcheck);
1866 if (rc_normal > 0) {
1867 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1868 file->private_data, iocb->ki_pos,
1870 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
1871 ktime_us_delta(ktime_get(), kstart));
1877 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1879 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1881 static int ll_file_get_iov_count(const struct iovec *iov,
1882 unsigned long *nr_segs, size_t *count,
1888 for (seg = 0; seg < *nr_segs; seg++) {
1889 const struct iovec *iv = &iov[seg];
1892 * If any segment has a negative length, or the cumulative
1893 * length ever wraps negative then return -EINVAL.
1896 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1898 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1903 cnt -= iv->iov_len; /* This segment is no good */
1910 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1911 unsigned long nr_segs, loff_t pos)
1918 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
1925 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1926 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
1927 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1928 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
1929 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1931 result = ll_file_read_iter(iocb, &to);
1936 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
1939 struct iovec iov = { .iov_base = buf, .iov_len = count };
1948 init_sync_kiocb(&kiocb, file);
1949 kiocb.ki_pos = *ppos;
1950 #ifdef HAVE_KIOCB_KI_LEFT
1951 kiocb.ki_left = count;
1952 #elif defined(HAVE_KI_NBYTES)
1953 kiocb.i_nbytes = count;
1956 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
1957 *ppos = kiocb.ki_pos;
1963 * Write to a file (through the page cache).
1966 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1967 unsigned long nr_segs, loff_t pos)
1969 struct iov_iter from;
1974 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
1981 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
1982 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
1983 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
1984 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
1985 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
1987 result = ll_file_write_iter(iocb, &from);
1992 static ssize_t ll_file_write(struct file *file, const char __user *buf,
1993 size_t count, loff_t *ppos)
1995 struct iovec iov = { .iov_base = (void __user *)buf,
2005 init_sync_kiocb(&kiocb, file);
2006 kiocb.ki_pos = *ppos;
2007 #ifdef HAVE_KIOCB_KI_LEFT
2008 kiocb.ki_left = count;
2009 #elif defined(HAVE_KI_NBYTES)
2010 kiocb.ki_nbytes = count;
2013 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2014 *ppos = kiocb.ki_pos;
2018 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2021 * Send file content (through pagecache) somewhere with helper
2023 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
2024 struct pipe_inode_info *pipe, size_t count,
2028 struct vvp_io_args *args;
2035 result = pcc_file_splice_read(in_file, ppos, pipe,
2036 count, flags, &cached);
2040 ll_ras_enter(in_file, *ppos, count);
2042 env = cl_env_get(&refcheck);
2044 RETURN(PTR_ERR(env));
2046 args = ll_env_args(env, IO_SPLICE);
2047 args->u.splice.via_pipe = pipe;
2048 args->u.splice.via_flags = flags;
2050 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
2051 cl_env_put(env, &refcheck);
2054 ll_rw_stats_tally(ll_i2sbi(file_inode(in_file)), current->pid,
2055 in_file->private_data, *ppos, result,
2060 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2061 __u64 flags, struct lov_user_md *lum, int lum_size)
2063 struct lookup_intent oit = {
2065 .it_flags = flags | MDS_OPEN_BY_FID,
2070 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2071 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2072 /* this code will only exist for big-endian systems */
2073 lustre_swab_lov_user_md(lum, 0);
2076 ll_inode_size_lock(inode);
2077 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2079 GOTO(out_unlock, rc);
2081 ll_release_openhandle(dentry, &oit);
2084 ll_inode_size_unlock(inode);
2085 ll_intent_release(&oit);
2090 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2091 struct lov_mds_md **lmmp, int *lmm_size,
2092 struct ptlrpc_request **request)
2094 struct ll_sb_info *sbi = ll_i2sbi(inode);
2095 struct mdt_body *body;
2096 struct lov_mds_md *lmm = NULL;
2097 struct ptlrpc_request *req = NULL;
2098 struct md_op_data *op_data;
2101 rc = ll_get_default_mdsize(sbi, &lmmsize);
2105 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2106 strlen(filename), lmmsize,
2107 LUSTRE_OPC_ANY, NULL);
2108 if (IS_ERR(op_data))
2109 RETURN(PTR_ERR(op_data));
2111 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2112 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2113 ll_finish_md_op_data(op_data);
2115 CDEBUG(D_INFO, "md_getattr_name failed "
2116 "on %s: rc %d\n", filename, rc);
2120 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2121 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2123 lmmsize = body->mbo_eadatasize;
2125 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2127 GOTO(out, rc = -ENODATA);
2130 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2131 LASSERT(lmm != NULL);
2133 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2134 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2135 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2136 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2137 GOTO(out, rc = -EPROTO);
2140 * This is coming from the MDS, so is probably in
2141 * little endian. We convert it to host endian before
2142 * passing it to userspace.
2144 if ((lmm->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
2145 __swab32(LOV_MAGIC_MAGIC)) {
2146 int stripe_count = 0;
2148 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2149 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2150 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2151 if (le32_to_cpu(lmm->lmm_pattern) &
2152 LOV_PATTERN_F_RELEASED)
2156 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2158 /* if function called for directory - we should
2159 * avoid swab not existent lsm objects */
2160 if (lmm->lmm_magic == LOV_MAGIC_V1 && S_ISREG(body->mbo_mode))
2161 lustre_swab_lov_user_md_objects(
2162 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2164 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2165 S_ISREG(body->mbo_mode))
2166 lustre_swab_lov_user_md_objects(
2167 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2173 *lmm_size = lmmsize;
2178 static int ll_lov_setea(struct inode *inode, struct file *file,
2181 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2182 struct lov_user_md *lump;
2183 int lum_size = sizeof(struct lov_user_md) +
2184 sizeof(struct lov_user_ost_data);
2188 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2191 OBD_ALLOC_LARGE(lump, lum_size);
2195 if (copy_from_user(lump, arg, lum_size))
2196 GOTO(out_lump, rc = -EFAULT);
2198 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2200 cl_lov_delay_create_clear(&file->f_flags);
2203 OBD_FREE_LARGE(lump, lum_size);
2207 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2214 env = cl_env_get(&refcheck);
2216 RETURN(PTR_ERR(env));
2218 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2219 cl_env_put(env, &refcheck);
2223 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2226 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2227 struct lov_user_md *klum;
2229 __u64 flags = FMODE_WRITE;
2232 rc = ll_copy_user_md(lum, &klum);
2237 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2242 rc = put_user(0, &lum->lmm_stripe_count);
2246 rc = ll_layout_refresh(inode, &gen);
2250 rc = ll_file_getstripe(inode, arg, lum_size);
2252 cl_lov_delay_create_clear(&file->f_flags);
2255 OBD_FREE_LARGE(klum, lum_size);
2261 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2263 struct ll_inode_info *lli = ll_i2info(inode);
2264 struct cl_object *obj = lli->lli_clob;
2265 struct ll_file_data *fd = file->private_data;
2266 struct ll_grouplock grouplock;
2271 CWARN("group id for group lock must not be 0\n");
2275 if (ll_file_nolock(file))
2276 RETURN(-EOPNOTSUPP);
2278 if (file->f_flags & O_NONBLOCK) {
2279 if (!mutex_trylock(&lli->lli_group_mutex))
2282 mutex_lock(&lli->lli_group_mutex);
2284 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2285 CWARN("group lock already existed with gid %lu\n",
2286 fd->fd_grouplock.lg_gid);
2287 GOTO(out, rc = -EINVAL);
2289 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2290 if (file->f_flags & O_NONBLOCK)
2291 GOTO(out, rc = -EAGAIN);
2292 mutex_unlock(&lli->lli_group_mutex);
2293 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2294 GOTO(retry, rc = 0);
2296 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2299 * XXX: group lock needs to protect all OST objects while PFL
2300 * can add new OST objects during the IO, so we'd instantiate
2301 * all OST objects before getting its group lock.
2306 struct cl_layout cl = {
2307 .cl_is_composite = false,
2309 struct lu_extent ext = {
2311 .e_end = OBD_OBJECT_EOF,
2314 env = cl_env_get(&refcheck);
2316 GOTO(out, rc = PTR_ERR(env));
2318 rc = cl_object_layout_get(env, obj, &cl);
2319 if (!rc && cl.cl_is_composite)
2320 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2323 cl_env_put(env, &refcheck);
2328 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2329 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2334 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2335 fd->fd_grouplock = grouplock;
2336 if (lli->lli_group_users == 0)
2337 lli->lli_group_gid = grouplock.lg_gid;
2338 lli->lli_group_users++;
2340 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2342 mutex_unlock(&lli->lli_group_mutex);
2347 static int ll_put_grouplock(struct inode *inode, struct file *file,
2350 struct ll_inode_info *lli = ll_i2info(inode);
2351 struct ll_file_data *fd = file->private_data;
2352 struct ll_grouplock grouplock;
2356 mutex_lock(&lli->lli_group_mutex);
2357 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2358 CWARN("no group lock held\n");
2359 GOTO(out, rc = -EINVAL);
2362 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2364 if (fd->fd_grouplock.lg_gid != arg) {
2365 CWARN("group lock %lu doesn't match current id %lu\n",
2366 arg, fd->fd_grouplock.lg_gid);
2367 GOTO(out, rc = -EINVAL);
2370 grouplock = fd->fd_grouplock;
2371 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2372 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2374 cl_put_grouplock(&grouplock);
2376 lli->lli_group_users--;
2377 if (lli->lli_group_users == 0) {
2378 lli->lli_group_gid = 0;
2379 wake_up_var(&lli->lli_group_users);
2381 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2384 mutex_unlock(&lli->lli_group_mutex);
2390 * Close inode open handle
2392 * \param dentry [in] dentry which contains the inode
2393 * \param it [in,out] intent which contains open info and result
2396 * \retval <0 failure
2398 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2400 struct inode *inode = dentry->d_inode;
2401 struct obd_client_handle *och;
2407 /* Root ? Do nothing. */
2408 if (dentry->d_inode->i_sb->s_root == dentry)
2411 /* No open handle to close? Move away */
2412 if (!it_disposition(it, DISP_OPEN_OPEN))
2415 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2417 OBD_ALLOC(och, sizeof(*och));
2419 GOTO(out, rc = -ENOMEM);
2421 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2425 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2427 /* this one is in place of ll_file_open */
2428 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2429 ptlrpc_req_finished(it->it_request);
2430 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2436 * Get size for inode for which FIEMAP mapping is requested.
2437 * Make the FIEMAP get_info call and returns the result.
2438 * \param fiemap kernel buffer to hold extens
2439 * \param num_bytes kernel buffer size
2441 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2447 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2450 /* Checks for fiemap flags */
2451 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2452 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2456 /* Check for FIEMAP_FLAG_SYNC */
2457 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2458 rc = filemap_fdatawrite(inode->i_mapping);
2463 env = cl_env_get(&refcheck);
2465 RETURN(PTR_ERR(env));
2467 if (i_size_read(inode) == 0) {
2468 rc = ll_glimpse_size(inode);
2473 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2474 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2475 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2477 /* If filesize is 0, then there would be no objects for mapping */
2478 if (fmkey.lfik_oa.o_size == 0) {
2479 fiemap->fm_mapped_extents = 0;
2483 fmkey.lfik_fiemap = *fiemap;
2485 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2486 &fmkey, fiemap, &num_bytes);
2488 cl_env_put(env, &refcheck);
2492 int ll_fid2path(struct inode *inode, void __user *arg)
2494 struct obd_export *exp = ll_i2mdexp(inode);
2495 const struct getinfo_fid2path __user *gfin = arg;
2497 struct getinfo_fid2path *gfout;
2503 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2504 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2507 /* Only need to get the buflen */
2508 if (get_user(pathlen, &gfin->gf_pathlen))
2511 if (pathlen > PATH_MAX)
2514 outsize = sizeof(*gfout) + pathlen;
2515 OBD_ALLOC(gfout, outsize);
2519 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2520 GOTO(gf_free, rc = -EFAULT);
2521 /* append root FID after gfout to let MDT know the root FID so that it
2522 * can lookup the correct path, this is mainly for fileset.
2523 * old server without fileset mount support will ignore this. */
2524 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2526 /* Call mdc_iocontrol */
2527 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2531 if (copy_to_user(arg, gfout, outsize))
2535 OBD_FREE(gfout, outsize);
2540 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2542 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2550 ioc->idv_version = 0;
2551 ioc->idv_layout_version = UINT_MAX;
2553 /* If no file object initialized, we consider its version is 0. */
2557 env = cl_env_get(&refcheck);
2559 RETURN(PTR_ERR(env));
2561 io = vvp_env_thread_io(env);
2563 io->u.ci_data_version.dv_data_version = 0;
2564 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2565 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2568 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2569 result = cl_io_loop(env, io);
2571 result = io->ci_result;
2573 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2574 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2576 cl_io_fini(env, io);
2578 if (unlikely(io->ci_need_restart))
2581 cl_env_put(env, &refcheck);
2587 * Read the data_version for inode.
2589 * This value is computed using stripe object version on OST.
2590 * Version is computed using server side locking.
2592 * @param flags if do sync on the OST side;
2594 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2595 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2597 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2599 struct ioc_data_version ioc = { .idv_flags = flags };
2602 rc = ll_ioc_data_version(inode, &ioc);
2604 *data_version = ioc.idv_version;
2610 * Trigger a HSM release request for the provided inode.
2612 int ll_hsm_release(struct inode *inode)
2615 struct obd_client_handle *och = NULL;
2616 __u64 data_version = 0;
2621 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2622 ll_i2sbi(inode)->ll_fsname,
2623 PFID(&ll_i2info(inode)->lli_fid));
2625 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2627 GOTO(out, rc = PTR_ERR(och));
2629 /* Grab latest data_version and [am]time values */
2630 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2634 env = cl_env_get(&refcheck);
2636 GOTO(out, rc = PTR_ERR(env));
2638 rc = ll_merge_attr(env, inode);
2639 cl_env_put(env, &refcheck);
2641 /* If error happen, we have the wrong size for a file.
2647 /* Release the file.
2648 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2649 * we still need it to pack l_remote_handle to MDT. */
2650 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2656 if (och != NULL && !IS_ERR(och)) /* close the file */
2657 ll_lease_close(och, inode, NULL);
2662 struct ll_swap_stack {
2665 struct inode *inode1;
2666 struct inode *inode2;
2671 static int ll_swap_layouts(struct file *file1, struct file *file2,
2672 struct lustre_swap_layouts *lsl)
2674 struct mdc_swap_layouts msl;
2675 struct md_op_data *op_data;
2678 struct ll_swap_stack *llss = NULL;
2681 OBD_ALLOC_PTR(llss);
2685 llss->inode1 = file_inode(file1);
2686 llss->inode2 = file_inode(file2);
2688 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2692 /* we use 2 bool because it is easier to swap than 2 bits */
2693 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2694 llss->check_dv1 = true;
2696 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2697 llss->check_dv2 = true;
2699 /* we cannot use lsl->sl_dvX directly because we may swap them */
2700 llss->dv1 = lsl->sl_dv1;
2701 llss->dv2 = lsl->sl_dv2;
2703 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2704 if (rc == 0) /* same file, done! */
2707 if (rc < 0) { /* sequentialize it */
2708 swap(llss->inode1, llss->inode2);
2710 swap(llss->dv1, llss->dv2);
2711 swap(llss->check_dv1, llss->check_dv2);
2715 if (gid != 0) { /* application asks to flush dirty cache */
2716 rc = ll_get_grouplock(llss->inode1, file1, gid);
2720 rc = ll_get_grouplock(llss->inode2, file2, gid);
2722 ll_put_grouplock(llss->inode1, file1, gid);
2727 /* ultimate check, before swaping the layouts we check if
2728 * dataversion has changed (if requested) */
2729 if (llss->check_dv1) {
2730 rc = ll_data_version(llss->inode1, &dv, 0);
2733 if (dv != llss->dv1)
2734 GOTO(putgl, rc = -EAGAIN);
2737 if (llss->check_dv2) {
2738 rc = ll_data_version(llss->inode2, &dv, 0);
2741 if (dv != llss->dv2)
2742 GOTO(putgl, rc = -EAGAIN);
2745 /* struct md_op_data is used to send the swap args to the mdt
2746 * only flags is missing, so we use struct mdc_swap_layouts
2747 * through the md_op_data->op_data */
2748 /* flags from user space have to be converted before they are send to
2749 * server, no flag is sent today, they are only used on the client */
2752 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2753 0, LUSTRE_OPC_ANY, &msl);
2754 if (IS_ERR(op_data))
2755 GOTO(free, rc = PTR_ERR(op_data));
2757 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2758 sizeof(*op_data), op_data, NULL);
2759 ll_finish_md_op_data(op_data);
2766 ll_put_grouplock(llss->inode2, file2, gid);
2767 ll_put_grouplock(llss->inode1, file1, gid);
2777 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2779 struct obd_export *exp = ll_i2mdexp(inode);
2780 struct md_op_data *op_data;
2784 /* Detect out-of range masks */
2785 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2788 /* Non-root users are forbidden to set or clear flags which are
2789 * NOT defined in HSM_USER_MASK. */
2790 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2791 !cfs_capable(CFS_CAP_SYS_ADMIN))
2794 if (!exp_connect_archive_id_array(exp)) {
2795 /* Detect out-of range archive id */
2796 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2797 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
2801 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2802 LUSTRE_OPC_ANY, hss);
2803 if (IS_ERR(op_data))
2804 RETURN(PTR_ERR(op_data));
2806 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
2809 ll_finish_md_op_data(op_data);
2814 static int ll_hsm_import(struct inode *inode, struct file *file,
2815 struct hsm_user_import *hui)
2817 struct hsm_state_set *hss = NULL;
2818 struct iattr *attr = NULL;
2822 if (!S_ISREG(inode->i_mode))
2828 GOTO(out, rc = -ENOMEM);
2830 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2831 hss->hss_archive_id = hui->hui_archive_id;
2832 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2833 rc = ll_hsm_state_set(inode, hss);
2837 OBD_ALLOC_PTR(attr);
2839 GOTO(out, rc = -ENOMEM);
2841 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2842 attr->ia_mode |= S_IFREG;
2843 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2844 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2845 attr->ia_size = hui->hui_size;
2846 attr->ia_mtime.tv_sec = hui->hui_mtime;
2847 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2848 attr->ia_atime.tv_sec = hui->hui_atime;
2849 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2851 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2852 ATTR_UID | ATTR_GID |
2853 ATTR_MTIME | ATTR_MTIME_SET |
2854 ATTR_ATIME | ATTR_ATIME_SET;
2858 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
2862 inode_unlock(inode);
2874 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2876 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2877 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2880 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
2882 struct inode *inode = file_inode(file);
2884 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
2885 ATTR_MTIME | ATTR_MTIME_SET |
2888 .tv_sec = lfu->lfu_atime_sec,
2889 .tv_nsec = lfu->lfu_atime_nsec,
2892 .tv_sec = lfu->lfu_mtime_sec,
2893 .tv_nsec = lfu->lfu_mtime_nsec,
2896 .tv_sec = lfu->lfu_ctime_sec,
2897 .tv_nsec = lfu->lfu_ctime_nsec,
2903 if (!capable(CAP_SYS_ADMIN))
2906 if (!S_ISREG(inode->i_mode))
2910 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
2912 inode_unlock(inode);
2917 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
2920 case MODE_READ_USER:
2922 case MODE_WRITE_USER:
2929 static const char *const user_lockname[] = LOCK_MODE_NAMES;
2931 /* Used to allow the upper layers of the client to request an LDLM lock
2932 * without doing an actual read or write.
2934 * Used for ladvise lockahead to manually request specific locks.
2936 * \param[in] file file this ladvise lock request is on
2937 * \param[in] ladvise ladvise struct describing this lock request
2939 * \retval 0 success, no detailed result available (sync requests
2940 * and requests sent to the server [not handled locally]
2941 * cannot return detailed results)
2942 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
2943 * see definitions for details.
2944 * \retval negative negative errno on error
2946 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
2948 struct lu_env *env = NULL;
2949 struct cl_io *io = NULL;
2950 struct cl_lock *lock = NULL;
2951 struct cl_lock_descr *descr = NULL;
2952 struct dentry *dentry = file->f_path.dentry;
2953 struct inode *inode = dentry->d_inode;
2954 enum cl_lock_mode cl_mode;
2955 off_t start = ladvise->lla_start;
2956 off_t end = ladvise->lla_end;
2962 CDEBUG(D_VFSTRACE, "Lock request: file=%.*s, inode=%p, mode=%s "
2963 "start=%llu, end=%llu\n", dentry->d_name.len,
2964 dentry->d_name.name, dentry->d_inode,
2965 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
2968 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
2970 GOTO(out, result = cl_mode);
2972 /* Get IO environment */
2973 result = cl_io_get(inode, &env, &io, &refcheck);
2977 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
2980 * nothing to do for this io. This currently happens when
2981 * stripe sub-object's are not yet created.
2983 result = io->ci_result;
2984 } else if (result == 0) {
2985 lock = vvp_env_lock(env);
2986 descr = &lock->cll_descr;
2988 descr->cld_obj = io->ci_obj;
2989 /* Convert byte offsets to pages */
2990 descr->cld_start = cl_index(io->ci_obj, start);
2991 descr->cld_end = cl_index(io->ci_obj, end);
2992 descr->cld_mode = cl_mode;
2993 /* CEF_MUST is used because we do not want to convert a
2994 * lockahead request to a lockless lock */
2995 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND |
2998 if (ladvise->lla_peradvice_flags & LF_ASYNC)
2999 descr->cld_enq_flags |= CEF_SPECULATIVE;
3001 result = cl_lock_request(env, io, lock);
3003 /* On success, we need to release the lock */
3005 cl_lock_release(env, lock);
3007 cl_io_fini(env, io);
3008 cl_env_put(env, &refcheck);
3010 /* -ECANCELED indicates a matching lock with a different extent
3011 * was already present, and -EEXIST indicates a matching lock
3012 * on exactly the same extent was already present.
3013 * We convert them to positive values for userspace to make
3014 * recognizing true errors easier.
3015 * Note we can only return these detailed results on async requests,
3016 * as sync requests look the same as i/o requests for locking. */
3017 if (result == -ECANCELED)
3018 result = LLA_RESULT_DIFFERENT;
3019 else if (result == -EEXIST)
3020 result = LLA_RESULT_SAME;
3025 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3027 static int ll_ladvise_sanity(struct inode *inode,
3028 struct llapi_lu_ladvise *ladvise)
3030 struct ll_sb_info *sbi = ll_i2sbi(inode);
3031 enum lu_ladvise_type advice = ladvise->lla_advice;
3032 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3033 * be in the first 32 bits of enum ladvise_flags */
3034 __u32 flags = ladvise->lla_peradvice_flags;
3035 /* 3 lines at 80 characters per line, should be plenty */
3038 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3041 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3042 sbi->ll_fsname, advice,
3043 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3047 /* Per-advice checks */
3049 case LU_LADVISE_LOCKNOEXPAND:
3050 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3052 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3053 "rc = %d\n", sbi->ll_fsname, flags,
3054 ladvise_names[advice], rc);
3058 case LU_LADVISE_LOCKAHEAD:
3059 /* Currently only READ and WRITE modes can be requested */
3060 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3061 ladvise->lla_lockahead_mode == 0) {
3063 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3064 "rc = %d\n", sbi->ll_fsname,
3065 ladvise->lla_lockahead_mode,
3066 ladvise_names[advice], rc);
3070 case LU_LADVISE_WILLREAD:
3071 case LU_LADVISE_DONTNEED:
3073 /* Note fall through above - These checks apply to all advices
3074 * except LOCKNOEXPAND */
3075 if (flags & ~LF_DEFAULT_MASK) {
3077 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3078 "rc = %d\n", sbi->ll_fsname, flags,
3079 ladvise_names[advice], rc);
3082 if (ladvise->lla_start >= ladvise->lla_end) {
3084 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3085 "for %s: rc = %d\n", sbi->ll_fsname,
3086 ladvise->lla_start, ladvise->lla_end,
3087 ladvise_names[advice], rc);
3099 * Give file access advices
3101 * The ladvise interface is similar to Linux fadvise() system call, except it
3102 * forwards the advices directly from Lustre client to server. The server side
3103 * codes will apply appropriate read-ahead and caching techniques for the
3104 * corresponding files.
3106 * A typical workload for ladvise is e.g. a bunch of different clients are
3107 * doing small random reads of a file, so prefetching pages into OSS cache
3108 * with big linear reads before the random IO is a net benefit. Fetching
3109 * all that data into each client cache with fadvise() may not be, due to
3110 * much more data being sent to the client.
3112 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3113 struct llapi_lu_ladvise *ladvise)
3117 struct cl_ladvise_io *lio;
3122 env = cl_env_get(&refcheck);
3124 RETURN(PTR_ERR(env));
3126 io = vvp_env_thread_io(env);
3127 io->ci_obj = ll_i2info(inode)->lli_clob;
3129 /* initialize parameters for ladvise */
3130 lio = &io->u.ci_ladvise;
3131 lio->li_start = ladvise->lla_start;
3132 lio->li_end = ladvise->lla_end;
3133 lio->li_fid = ll_inode2fid(inode);
3134 lio->li_advice = ladvise->lla_advice;
3135 lio->li_flags = flags;
3137 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3138 rc = cl_io_loop(env, io);
3142 cl_io_fini(env, io);
3143 cl_env_put(env, &refcheck);
3147 static int ll_lock_noexpand(struct file *file, int flags)
3149 struct ll_file_data *fd = file->private_data;
3151 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3156 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3159 struct fsxattr fsxattr;
3161 if (copy_from_user(&fsxattr,
3162 (const struct fsxattr __user *)arg,
3166 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3167 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
3168 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3169 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3170 if (copy_to_user((struct fsxattr __user *)arg,
3171 &fsxattr, sizeof(fsxattr)))
3177 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa)
3180 * Project Quota ID state is only allowed to change from within the init
3181 * namespace. Enforce that restriction only if we are trying to change
3182 * the quota ID state. Everything else is allowed in user namespaces.
3184 if (current_user_ns() == &init_user_ns)
3187 if (ll_i2info(inode)->lli_projid != fa->fsx_projid)
3190 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT)) {
3191 if (!(fa->fsx_xflags & FS_XFLAG_PROJINHERIT))
3194 if (fa->fsx_xflags & FS_XFLAG_PROJINHERIT)
3201 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3205 struct md_op_data *op_data;
3206 struct ptlrpc_request *req = NULL;
3208 struct fsxattr fsxattr;
3209 struct cl_object *obj;
3213 if (copy_from_user(&fsxattr,
3214 (const struct fsxattr __user *)arg,
3218 rc = ll_ioctl_check_project(inode, &fsxattr);
3222 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3223 LUSTRE_OPC_ANY, NULL);
3224 if (IS_ERR(op_data))
3225 RETURN(PTR_ERR(op_data));
3227 flags = ll_xflags_to_inode_flags(fsxattr.fsx_xflags);
3228 op_data->op_attr_flags = ll_inode_to_ext_flags(flags);
3229 if (fsxattr.fsx_xflags & FS_XFLAG_PROJINHERIT)
3230 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3231 op_data->op_projid = fsxattr.fsx_projid;
3232 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3233 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL,
3235 ptlrpc_req_finished(req);
3237 GOTO(out_fsxattr, rc);
3238 ll_update_inode_flags(inode, op_data->op_attr_flags);
3239 obj = ll_i2info(inode)->lli_clob;
3241 GOTO(out_fsxattr, rc);
3243 /* Avoiding OST RPC if this is only project ioctl */
3244 if (fsxattr.fsx_xflags == 0 ||
3245 fsxattr.fsx_xflags == FS_XFLAG_PROJINHERIT)
3246 GOTO(out_fsxattr, rc);
3248 OBD_ALLOC_PTR(attr);
3250 GOTO(out_fsxattr, rc = -ENOMEM);
3252 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS,
3253 fsxattr.fsx_xflags);
3256 ll_finish_md_op_data(op_data);
3260 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3263 struct inode *inode = file_inode(file);
3264 struct ll_file_data *fd = file->private_data;
3265 struct ll_inode_info *lli = ll_i2info(inode);
3266 struct obd_client_handle *och = NULL;
3267 struct split_param sp;
3268 struct pcc_param param;
3269 bool lease_broken = false;
3271 enum mds_op_bias bias = 0;
3272 struct file *layout_file = NULL;
3274 size_t data_size = 0;
3275 bool attached = false;
3280 mutex_lock(&lli->lli_och_mutex);
3281 if (fd->fd_lease_och != NULL) {
3282 och = fd->fd_lease_och;
3283 fd->fd_lease_och = NULL;
3285 mutex_unlock(&lli->lli_och_mutex);
3290 fmode = och->och_flags;
3292 switch (ioc->lil_flags) {
3293 case LL_LEASE_RESYNC_DONE:
3294 if (ioc->lil_count > IOC_IDS_MAX)
3295 GOTO(out_lease_close, rc = -EINVAL);
3297 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3298 OBD_ALLOC(data, data_size);
3300 GOTO(out_lease_close, rc = -ENOMEM);
3302 if (copy_from_user(data, (void __user *)arg, data_size))
3303 GOTO(out_lease_close, rc = -EFAULT);
3305 bias = MDS_CLOSE_RESYNC_DONE;
3307 case LL_LEASE_LAYOUT_MERGE: {
3310 if (ioc->lil_count != 1)
3311 GOTO(out_lease_close, rc = -EINVAL);
3313 arg += sizeof(*ioc);
3314 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3315 GOTO(out_lease_close, rc = -EFAULT);
3317 layout_file = fget(fd);
3319 GOTO(out_lease_close, rc = -EBADF);
3321 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3322 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3323 GOTO(out_lease_close, rc = -EPERM);
3325 data = file_inode(layout_file);
3326 bias = MDS_CLOSE_LAYOUT_MERGE;
3329 case LL_LEASE_LAYOUT_SPLIT: {
3333 if (ioc->lil_count != 2)
3334 GOTO(out_lease_close, rc = -EINVAL);
3336 arg += sizeof(*ioc);
3337 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3338 GOTO(out_lease_close, rc = -EFAULT);
3340 arg += sizeof(__u32);
3341 if (copy_from_user(&mirror_id, (void __user *)arg,
3343 GOTO(out_lease_close, rc = -EFAULT);
3345 layout_file = fget(fdv);
3347 GOTO(out_lease_close, rc = -EBADF);
3349 sp.sp_inode = file_inode(layout_file);
3350 sp.sp_mirror_id = (__u16)mirror_id;
3352 bias = MDS_CLOSE_LAYOUT_SPLIT;
3355 case LL_LEASE_PCC_ATTACH:
3356 if (ioc->lil_count != 1)
3359 arg += sizeof(*ioc);
3360 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3362 GOTO(out_lease_close, rc2 = -EFAULT);
3364 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3366 GOTO(out_lease_close, rc2);
3369 /* Grab latest data version */
3370 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3373 GOTO(out_lease_close, rc2);
3376 bias = MDS_PCC_ATTACH;
3379 /* without close intent */
3384 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3388 rc = ll_lease_och_release(inode, file);
3397 switch (ioc->lil_flags) {
3398 case LL_LEASE_RESYNC_DONE:
3400 OBD_FREE(data, data_size);
3402 case LL_LEASE_LAYOUT_MERGE:
3403 case LL_LEASE_LAYOUT_SPLIT:
3407 case LL_LEASE_PCC_ATTACH:
3410 rc = pcc_readwrite_attach_fini(file, inode,
3411 param.pa_layout_gen,
3418 rc = ll_lease_type_from_fmode(fmode);
3422 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3425 struct inode *inode = file_inode(file);
3426 struct ll_inode_info *lli = ll_i2info(inode);
3427 struct ll_file_data *fd = file->private_data;
3428 struct obd_client_handle *och = NULL;
3429 __u64 open_flags = 0;
3435 switch (ioc->lil_mode) {
3436 case LL_LEASE_WRLCK:
3437 if (!(file->f_mode & FMODE_WRITE))
3439 fmode = FMODE_WRITE;
3441 case LL_LEASE_RDLCK:
3442 if (!(file->f_mode & FMODE_READ))
3446 case LL_LEASE_UNLCK:
3447 RETURN(ll_file_unlock_lease(file, ioc, arg));
3452 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3454 /* apply for lease */
3455 if (ioc->lil_flags & LL_LEASE_RESYNC)
3456 open_flags = MDS_OPEN_RESYNC;
3457 och = ll_lease_open(inode, file, fmode, open_flags);
3459 RETURN(PTR_ERR(och));
3461 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3462 rc = ll_lease_file_resync(och, inode, arg);
3464 ll_lease_close(och, inode, NULL);
3467 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3469 ll_lease_close(och, inode, NULL);
3475 mutex_lock(&lli->lli_och_mutex);
3476 if (fd->fd_lease_och == NULL) {
3477 fd->fd_lease_och = och;
3480 mutex_unlock(&lli->lli_och_mutex);
3482 /* impossible now that only excl is supported for now */
3483 ll_lease_close(och, inode, &lease_broken);
3489 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3491 struct ll_inode_info *lli = ll_i2info(inode);
3492 struct ll_sb_info *sbi = ll_i2sbi(inode);
3493 __u64 now = ktime_get_real_seconds();
3496 spin_lock(&lli->lli_heat_lock);
3497 heat->lh_flags = lli->lli_heat_flags;
3498 for (i = 0; i < heat->lh_count; i++)
3499 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3500 now, sbi->ll_heat_decay_weight,
3501 sbi->ll_heat_period_second);
3502 spin_unlock(&lli->lli_heat_lock);
3505 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3507 struct ll_inode_info *lli = ll_i2info(inode);
3510 spin_lock(&lli->lli_heat_lock);
3511 if (flags & LU_HEAT_FLAG_CLEAR)
3512 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3514 if (flags & LU_HEAT_FLAG_OFF)
3515 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3517 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3519 spin_unlock(&lli->lli_heat_lock);
3525 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3527 struct inode *inode = file_inode(file);
3528 struct ll_file_data *fd = file->private_data;
3532 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3533 PFID(ll_inode2fid(inode)), inode, cmd);
3534 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3536 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3537 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3541 case LL_IOC_GETFLAGS:
3542 /* Get the current value of the file flags */
3543 return put_user(fd->fd_flags, (int __user *)arg);
3544 case LL_IOC_SETFLAGS:
3545 case LL_IOC_CLRFLAGS:
3546 /* Set or clear specific file flags */
3547 /* XXX This probably needs checks to ensure the flags are
3548 * not abused, and to handle any flag side effects.
3550 if (get_user(flags, (int __user *) arg))
3553 if (cmd == LL_IOC_SETFLAGS) {
3554 if ((flags & LL_FILE_IGNORE_LOCK) &&
3555 !(file->f_flags & O_DIRECT)) {
3556 CERROR("%s: unable to disable locking on "
3557 "non-O_DIRECT file\n", current->comm);
3561 fd->fd_flags |= flags;
3563 fd->fd_flags &= ~flags;
3566 case LL_IOC_LOV_SETSTRIPE:
3567 case LL_IOC_LOV_SETSTRIPE_NEW:
3568 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3569 case LL_IOC_LOV_SETEA:
3570 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3571 case LL_IOC_LOV_SWAP_LAYOUTS: {
3573 struct lustre_swap_layouts lsl;
3575 if (copy_from_user(&lsl, (char __user *)arg,
3576 sizeof(struct lustre_swap_layouts)))
3579 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3582 file2 = fget(lsl.sl_fd);
3586 /* O_WRONLY or O_RDWR */
3587 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3588 GOTO(out, rc = -EPERM);
3590 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3591 struct inode *inode2;
3592 struct ll_inode_info *lli;
3593 struct obd_client_handle *och = NULL;
3595 lli = ll_i2info(inode);
3596 mutex_lock(&lli->lli_och_mutex);
3597 if (fd->fd_lease_och != NULL) {
3598 och = fd->fd_lease_och;
3599 fd->fd_lease_och = NULL;
3601 mutex_unlock(&lli->lli_och_mutex);
3603 GOTO(out, rc = -ENOLCK);
3604 inode2 = file_inode(file2);
3605 rc = ll_swap_layouts_close(och, inode, inode2);
3607 rc = ll_swap_layouts(file, file2, &lsl);
3613 case LL_IOC_LOV_GETSTRIPE:
3614 case LL_IOC_LOV_GETSTRIPE_NEW:
3615 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
3616 case FS_IOC_GETFLAGS:
3617 case FS_IOC_SETFLAGS:
3618 RETURN(ll_iocontrol(inode, file, cmd, arg));
3619 case FSFILT_IOC_GETVERSION:
3620 case FS_IOC_GETVERSION:
3621 RETURN(put_user(inode->i_generation, (int __user *)arg));
3622 /* We need to special case any other ioctls we want to handle,
3623 * to send them to the MDS/OST as appropriate and to properly
3624 * network encode the arg field. */
3625 case FS_IOC_SETVERSION:
3628 case LL_IOC_GROUP_LOCK:
3629 RETURN(ll_get_grouplock(inode, file, arg));
3630 case LL_IOC_GROUP_UNLOCK:
3631 RETURN(ll_put_grouplock(inode, file, arg));
3632 case IOC_OBD_STATFS:
3633 RETURN(ll_obd_statfs(inode, (void __user *)arg));
3635 case LL_IOC_FLUSHCTX:
3636 RETURN(ll_flush_ctx(inode));
3637 case LL_IOC_PATH2FID: {
3638 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
3639 sizeof(struct lu_fid)))
3644 case LL_IOC_GETPARENT:
3645 RETURN(ll_getparent(file, (struct getparent __user *)arg));
3647 case OBD_IOC_FID2PATH:
3648 RETURN(ll_fid2path(inode, (void __user *)arg));
3649 case LL_IOC_DATA_VERSION: {
3650 struct ioc_data_version idv;
3653 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
3656 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
3657 rc = ll_ioc_data_version(inode, &idv);
3660 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
3666 case LL_IOC_GET_MDTIDX: {
3669 mdtidx = ll_get_mdt_idx(inode);
3673 if (put_user((int)mdtidx, (int __user *)arg))
3678 case OBD_IOC_GETDTNAME:
3679 case OBD_IOC_GETMDNAME:
3680 RETURN(ll_get_obd_name(inode, cmd, arg));
3681 case LL_IOC_HSM_STATE_GET: {
3682 struct md_op_data *op_data;
3683 struct hsm_user_state *hus;
3690 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3691 LUSTRE_OPC_ANY, hus);
3692 if (IS_ERR(op_data)) {
3694 RETURN(PTR_ERR(op_data));
3697 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3700 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
3703 ll_finish_md_op_data(op_data);
3707 case LL_IOC_HSM_STATE_SET: {
3708 struct hsm_state_set *hss;
3715 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
3720 rc = ll_hsm_state_set(inode, hss);
3725 case LL_IOC_HSM_ACTION: {
3726 struct md_op_data *op_data;
3727 struct hsm_current_action *hca;
3734 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3735 LUSTRE_OPC_ANY, hca);
3736 if (IS_ERR(op_data)) {
3738 RETURN(PTR_ERR(op_data));
3741 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3744 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
3747 ll_finish_md_op_data(op_data);
3751 case LL_IOC_SET_LEASE_OLD: {
3752 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
3754 RETURN(ll_file_set_lease(file, &ioc, 0));
3756 case LL_IOC_SET_LEASE: {
3757 struct ll_ioc_lease ioc;
3759 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
3762 RETURN(ll_file_set_lease(file, &ioc, arg));
3764 case LL_IOC_GET_LEASE: {
3765 struct ll_inode_info *lli = ll_i2info(inode);
3766 struct ldlm_lock *lock = NULL;
3769 mutex_lock(&lli->lli_och_mutex);
3770 if (fd->fd_lease_och != NULL) {
3771 struct obd_client_handle *och = fd->fd_lease_och;
3773 lock = ldlm_handle2lock(&och->och_lease_handle);
3775 lock_res_and_lock(lock);
3776 if (!ldlm_is_cancel(lock))
3777 fmode = och->och_flags;
3779 unlock_res_and_lock(lock);
3780 LDLM_LOCK_PUT(lock);
3783 mutex_unlock(&lli->lli_och_mutex);
3785 RETURN(ll_lease_type_from_fmode(fmode));
3787 case LL_IOC_HSM_IMPORT: {
3788 struct hsm_user_import *hui;
3794 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
3799 rc = ll_hsm_import(inode, file, hui);
3804 case LL_IOC_FUTIMES_3: {
3805 struct ll_futimes_3 lfu;
3807 if (copy_from_user(&lfu,
3808 (const struct ll_futimes_3 __user *)arg,
3812 RETURN(ll_file_futimes_3(file, &lfu));
3814 case LL_IOC_LADVISE: {
3815 struct llapi_ladvise_hdr *k_ladvise_hdr;
3816 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
3819 int alloc_size = sizeof(*k_ladvise_hdr);
3822 u_ladvise_hdr = (void __user *)arg;
3823 OBD_ALLOC_PTR(k_ladvise_hdr);
3824 if (k_ladvise_hdr == NULL)
3827 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3828 GOTO(out_ladvise, rc = -EFAULT);
3830 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
3831 k_ladvise_hdr->lah_count < 1)
3832 GOTO(out_ladvise, rc = -EINVAL);
3834 num_advise = k_ladvise_hdr->lah_count;
3835 if (num_advise >= LAH_COUNT_MAX)
3836 GOTO(out_ladvise, rc = -EFBIG);
3838 OBD_FREE_PTR(k_ladvise_hdr);
3839 alloc_size = offsetof(typeof(*k_ladvise_hdr),
3840 lah_advise[num_advise]);
3841 OBD_ALLOC(k_ladvise_hdr, alloc_size);
3842 if (k_ladvise_hdr == NULL)
3846 * TODO: submit multiple advices to one server in a single RPC
3848 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3849 GOTO(out_ladvise, rc = -EFAULT);
3851 for (i = 0; i < num_advise; i++) {
3852 struct llapi_lu_ladvise *k_ladvise =
3853 &k_ladvise_hdr->lah_advise[i];
3854 struct llapi_lu_ladvise __user *u_ladvise =
3855 &u_ladvise_hdr->lah_advise[i];
3857 rc = ll_ladvise_sanity(inode, k_ladvise);
3859 GOTO(out_ladvise, rc);
3861 switch (k_ladvise->lla_advice) {
3862 case LU_LADVISE_LOCKNOEXPAND:
3863 rc = ll_lock_noexpand(file,
3864 k_ladvise->lla_peradvice_flags);
3865 GOTO(out_ladvise, rc);
3866 case LU_LADVISE_LOCKAHEAD:
3868 rc = ll_file_lock_ahead(file, k_ladvise);
3871 GOTO(out_ladvise, rc);
3874 &u_ladvise->lla_lockahead_result))
3875 GOTO(out_ladvise, rc = -EFAULT);
3878 rc = ll_ladvise(inode, file,
3879 k_ladvise_hdr->lah_flags,
3882 GOTO(out_ladvise, rc);
3889 OBD_FREE(k_ladvise_hdr, alloc_size);
3892 case LL_IOC_FLR_SET_MIRROR: {
3893 /* mirror I/O must be direct to avoid polluting page cache
3895 if (!(file->f_flags & O_DIRECT))
3898 fd->fd_designated_mirror = (__u32)arg;
3901 case LL_IOC_FSGETXATTR:
3902 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
3903 case LL_IOC_FSSETXATTR:
3904 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
3906 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
3907 case LL_IOC_HEAT_GET: {
3908 struct lu_heat uheat;
3909 struct lu_heat *heat;
3912 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
3915 if (uheat.lh_count > OBD_HEAT_COUNT)
3916 uheat.lh_count = OBD_HEAT_COUNT;
3918 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
3919 OBD_ALLOC(heat, size);
3923 heat->lh_count = uheat.lh_count;
3924 ll_heat_get(inode, heat);
3925 rc = copy_to_user((char __user *)arg, heat, size);
3926 OBD_FREE(heat, size);
3927 RETURN(rc ? -EFAULT : 0);
3929 case LL_IOC_HEAT_SET: {
3932 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
3935 rc = ll_heat_set(inode, flags);
3938 case LL_IOC_PCC_DETACH: {
3939 struct lu_pcc_detach *detach;
3941 OBD_ALLOC_PTR(detach);
3945 if (copy_from_user(detach,
3946 (const struct lu_pcc_detach __user *)arg,
3948 GOTO(out_detach_free, rc = -EFAULT);
3950 if (!S_ISREG(inode->i_mode))
3951 GOTO(out_detach_free, rc = -EINVAL);
3953 if (!inode_owner_or_capable(inode))
3954 GOTO(out_detach_free, rc = -EPERM);
3956 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
3958 OBD_FREE_PTR(detach);
3961 case LL_IOC_PCC_STATE: {
3962 struct lu_pcc_state __user *ustate =
3963 (struct lu_pcc_state __user *)arg;
3964 struct lu_pcc_state *state;
3966 OBD_ALLOC_PTR(state);
3970 if (copy_from_user(state, ustate, sizeof(*state)))
3971 GOTO(out_state, rc = -EFAULT);
3973 rc = pcc_ioctl_state(file, inode, state);
3975 GOTO(out_state, rc);
3977 if (copy_to_user(ustate, state, sizeof(*state)))
3978 GOTO(out_state, rc = -EFAULT);
3981 OBD_FREE_PTR(state);
3985 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
3986 (void __user *)arg));
3990 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
3992 struct inode *inode = file_inode(file);
3993 loff_t retval, eof = 0;
3994 ktime_t kstart = ktime_get();
3997 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
3998 (origin == SEEK_CUR) ? file->f_pos : 0);
3999 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4000 PFID(ll_inode2fid(inode)), inode, retval, retval,
4003 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
4004 retval = ll_glimpse_size(inode);
4007 eof = i_size_read(inode);
4010 retval = generic_file_llseek_size(file, offset, origin,
4011 ll_file_maxbytes(inode), eof);
4013 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4014 ktime_us_delta(ktime_get(), kstart));
4018 static int ll_flush(struct file *file, fl_owner_t id)
4020 struct inode *inode = file_inode(file);
4021 struct ll_inode_info *lli = ll_i2info(inode);
4022 struct ll_file_data *fd = file->private_data;
4025 LASSERT(!S_ISDIR(inode->i_mode));
4027 /* catch async errors that were recorded back when async writeback
4028 * failed for pages in this mapping. */
4029 rc = lli->lli_async_rc;
4030 lli->lli_async_rc = 0;
4031 if (lli->lli_clob != NULL) {
4032 err = lov_read_and_clear_async_rc(lli->lli_clob);
4037 /* The application has been told write failure already.
4038 * Do not report failure again. */
4039 if (fd->fd_write_failed)
4041 return rc ? -EIO : 0;
4045 * Called to make sure a portion of file has been written out.
4046 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4048 * Return how many pages have been written.
4050 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4051 enum cl_fsync_mode mode, int ignore_layout)
4055 struct cl_fsync_io *fio;
4060 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4061 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4064 env = cl_env_get(&refcheck);
4066 RETURN(PTR_ERR(env));
4068 io = vvp_env_thread_io(env);
4069 io->ci_obj = ll_i2info(inode)->lli_clob;
4070 io->ci_ignore_layout = ignore_layout;
4072 /* initialize parameters for sync */
4073 fio = &io->u.ci_fsync;
4074 fio->fi_start = start;
4076 fio->fi_fid = ll_inode2fid(inode);
4077 fio->fi_mode = mode;
4078 fio->fi_nr_written = 0;
4080 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4081 result = cl_io_loop(env, io);
4083 result = io->ci_result;
4085 result = fio->fi_nr_written;
4086 cl_io_fini(env, io);
4087 cl_env_put(env, &refcheck);
4093 * When dentry is provided (the 'else' case), file_dentry() may be
4094 * null and dentry must be used directly rather than pulled from
4095 * file_dentry() as is done otherwise.
4098 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4100 struct dentry *dentry = file_dentry(file);
4101 struct inode *inode = dentry->d_inode;
4102 struct ll_inode_info *lli = ll_i2info(inode);
4103 struct ptlrpc_request *req;
4104 ktime_t kstart = ktime_get();
4110 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4111 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4113 /* fsync's caller has already called _fdata{sync,write}, we want
4114 * that IO to finish before calling the osc and mdc sync methods */
4115 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4118 /* catch async errors that were recorded back when async writeback
4119 * failed for pages in this mapping. */
4120 if (!S_ISDIR(inode->i_mode)) {
4121 err = lli->lli_async_rc;
4122 lli->lli_async_rc = 0;
4125 if (lli->lli_clob != NULL) {
4126 err = lov_read_and_clear_async_rc(lli->lli_clob);
4132 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4136 ptlrpc_req_finished(req);
4138 if (S_ISREG(inode->i_mode)) {
4139 struct ll_file_data *fd = file->private_data;
4142 /* Sync metadata on MDT first, and then sync the cached data
4145 err = pcc_fsync(file, start, end, datasync, &cached);
4147 err = cl_sync_file_range(inode, start, end,
4149 if (rc == 0 && err < 0)
4152 fd->fd_write_failed = true;
4154 fd->fd_write_failed = false;
4157 inode_unlock(inode);
4160 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4161 ktime_us_delta(ktime_get(), kstart));
4166 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4168 struct inode *inode = file_inode(file);
4169 struct ll_sb_info *sbi = ll_i2sbi(inode);
4170 struct ldlm_enqueue_info einfo = {
4171 .ei_type = LDLM_FLOCK,
4172 .ei_cb_cp = ldlm_flock_completion_ast,
4173 .ei_cbdata = file_lock,
4175 struct md_op_data *op_data;
4176 struct lustre_handle lockh = { 0 };
4177 union ldlm_policy_data flock = { { 0 } };
4178 int fl_type = file_lock->fl_type;
4179 ktime_t kstart = ktime_get();
4185 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4186 PFID(ll_inode2fid(inode)), file_lock);
4188 if (file_lock->fl_flags & FL_FLOCK) {
4189 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4190 /* flocks are whole-file locks */
4191 flock.l_flock.end = OFFSET_MAX;
4192 /* For flocks owner is determined by the local file desctiptor*/
4193 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4194 } else if (file_lock->fl_flags & FL_POSIX) {
4195 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4196 flock.l_flock.start = file_lock->fl_start;
4197 flock.l_flock.end = file_lock->fl_end;
4201 flock.l_flock.pid = file_lock->fl_pid;
4203 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4204 /* Somewhat ugly workaround for svc lockd.
4205 * lockd installs custom fl_lmops->lm_compare_owner that checks
4206 * for the fl_owner to be the same (which it always is on local node
4207 * I guess between lockd processes) and then compares pid.
4208 * As such we assign pid to the owner field to make it all work,
4209 * conflict with normal locks is unlikely since pid space and
4210 * pointer space for current->files are not intersecting */
4211 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4212 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4217 einfo.ei_mode = LCK_PR;
4220 /* An unlock request may or may not have any relation to
4221 * existing locks so we may not be able to pass a lock handle
4222 * via a normal ldlm_lock_cancel() request. The request may even
4223 * unlock a byte range in the middle of an existing lock. In
4224 * order to process an unlock request we need all of the same
4225 * information that is given with a normal read or write record
4226 * lock request. To avoid creating another ldlm unlock (cancel)
4227 * message we'll treat a LCK_NL flock request as an unlock. */
4228 einfo.ei_mode = LCK_NL;
4231 einfo.ei_mode = LCK_PW;
4234 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4249 flags = LDLM_FL_BLOCK_NOWAIT;
4255 flags = LDLM_FL_TEST_LOCK;
4258 CERROR("unknown fcntl lock command: %d\n", cmd);
4262 /* Save the old mode so that if the mode in the lock changes we
4263 * can decrement the appropriate reader or writer refcount. */
4264 file_lock->fl_type = einfo.ei_mode;
4266 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4267 LUSTRE_OPC_ANY, NULL);
4268 if (IS_ERR(op_data))
4269 RETURN(PTR_ERR(op_data));
4271 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4272 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4273 flock.l_flock.pid, flags, einfo.ei_mode,
4274 flock.l_flock.start, flock.l_flock.end);
4276 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4279 /* Restore the file lock type if not TEST lock. */
4280 if (!(flags & LDLM_FL_TEST_LOCK))
4281 file_lock->fl_type = fl_type;
4283 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4284 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4285 !(flags & LDLM_FL_TEST_LOCK))
4286 rc2 = locks_lock_file_wait(file, file_lock);
4288 if ((file_lock->fl_flags & FL_FLOCK) &&
4289 (rc == 0 || file_lock->fl_type == F_UNLCK))
4290 rc2 = flock_lock_file_wait(file, file_lock);
4291 if ((file_lock->fl_flags & FL_POSIX) &&
4292 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4293 !(flags & LDLM_FL_TEST_LOCK))
4294 rc2 = posix_lock_file_wait(file, file_lock);
4295 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4297 if (rc2 && file_lock->fl_type != F_UNLCK) {
4298 einfo.ei_mode = LCK_NL;
4299 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4304 ll_finish_md_op_data(op_data);
4307 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
4308 ktime_us_delta(ktime_get(), kstart));
4312 int ll_get_fid_by_name(struct inode *parent, const char *name,
4313 int namelen, struct lu_fid *fid,
4314 struct inode **inode)
4316 struct md_op_data *op_data = NULL;
4317 struct mdt_body *body;
4318 struct ptlrpc_request *req;
4322 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4323 LUSTRE_OPC_ANY, NULL);
4324 if (IS_ERR(op_data))
4325 RETURN(PTR_ERR(op_data));
4327 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4328 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4329 ll_finish_md_op_data(op_data);
4333 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4335 GOTO(out_req, rc = -EFAULT);
4337 *fid = body->mbo_fid1;
4340 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
4342 ptlrpc_req_finished(req);
4346 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4349 struct dentry *dchild = NULL;
4350 struct inode *child_inode = NULL;
4351 struct md_op_data *op_data;
4352 struct ptlrpc_request *request = NULL;
4353 struct obd_client_handle *och = NULL;
4355 struct mdt_body *body;
4356 __u64 data_version = 0;
4357 size_t namelen = strlen(name);
4358 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4362 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4363 PFID(ll_inode2fid(parent)), name,
4364 lum->lum_stripe_offset, lum->lum_stripe_count);
4366 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4367 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4368 lustre_swab_lmv_user_md(lum);
4370 /* Get child FID first */
4371 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4374 dchild = d_lookup(file_dentry(file), &qstr);
4376 if (dchild->d_inode)
4377 child_inode = igrab(dchild->d_inode);
4382 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4391 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4392 OBD_CONNECT2_DIR_MIGRATE)) {
4393 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4394 ll_dir_striped(child_inode)) {
4395 CERROR("%s: MDT doesn't support stripe directory "
4396 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4397 GOTO(out_iput, rc = -EOPNOTSUPP);
4402 * lfs migrate command needs to be blocked on the client
4403 * by checking the migrate FID against the FID of the
4406 if (child_inode == parent->i_sb->s_root->d_inode)
4407 GOTO(out_iput, rc = -EINVAL);
4409 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4410 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
4411 if (IS_ERR(op_data))
4412 GOTO(out_iput, rc = PTR_ERR(op_data));
4414 inode_lock(child_inode);
4415 op_data->op_fid3 = *ll_inode2fid(child_inode);
4416 if (!fid_is_sane(&op_data->op_fid3)) {
4417 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4418 ll_i2sbi(parent)->ll_fsname, name,
4419 PFID(&op_data->op_fid3));
4420 GOTO(out_unlock, rc = -EINVAL);
4423 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4424 op_data->op_data = lum;
4425 op_data->op_data_size = lumlen;
4428 if (S_ISREG(child_inode->i_mode)) {
4429 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4433 GOTO(out_unlock, rc);
4436 rc = ll_data_version(child_inode, &data_version,
4439 GOTO(out_close, rc);
4441 op_data->op_open_handle = och->och_open_handle;
4442 op_data->op_data_version = data_version;
4443 op_data->op_lease_handle = och->och_lease_handle;
4444 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4446 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4447 och->och_mod->mod_open_req->rq_replay = 0;
4448 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4451 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name, namelen,
4452 name, namelen, &request);
4454 LASSERT(request != NULL);
4455 ll_update_times(request, parent);
4458 if (rc == 0 || rc == -EAGAIN) {
4459 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4460 LASSERT(body != NULL);
4462 /* If the server does release layout lock, then we cleanup
4463 * the client och here, otherwise release it in out_close: */
4464 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4465 obd_mod_put(och->och_mod);
4466 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4468 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4474 if (request != NULL) {
4475 ptlrpc_req_finished(request);
4479 /* Try again if the lease has cancelled. */
4480 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4485 ll_lease_close(och, child_inode, NULL);
4487 clear_nlink(child_inode);
4489 inode_unlock(child_inode);
4490 ll_finish_md_op_data(op_data);
4497 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
4499 struct ll_file_data *fd = file->private_data;
4503 * In order to avoid flood of warning messages, only print one message
4504 * for one file. And the entire message rate on the client is limited
4505 * by CDEBUG_LIMIT too.
4507 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
4508 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
4509 CDEBUG_LIMIT(D_TTY | D_CONSOLE,
4510 "flock disabled, mount with '-o [local]flock' to enable\r\n");
4516 * test if some locks matching bits and l_req_mode are acquired
4517 * - bits can be in different locks
4518 * - if found clear the common lock bits in *bits
4519 * - the bits not found, are kept in *bits
4521 * \param bits [IN] searched lock bits [IN]
4522 * \param l_req_mode [IN] searched lock mode
4523 * \retval boolean, true iff all bits are found
4525 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
4527 struct lustre_handle lockh;
4528 union ldlm_policy_data policy;
4529 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
4530 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
4539 fid = &ll_i2info(inode)->lli_fid;
4540 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
4541 ldlm_lockname[mode]);
4543 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
4544 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
4545 policy.l_inodebits.bits = *bits & (1 << i);
4546 if (policy.l_inodebits.bits == 0)
4549 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
4550 &policy, mode, &lockh)) {
4551 struct ldlm_lock *lock;
4553 lock = ldlm_handle2lock(&lockh);
4556 ~(lock->l_policy_data.l_inodebits.bits);
4557 LDLM_LOCK_PUT(lock);
4559 *bits &= ~policy.l_inodebits.bits;
4566 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
4567 struct lustre_handle *lockh, __u64 flags,
4568 enum ldlm_mode mode)
4570 union ldlm_policy_data policy = { .l_inodebits = { bits } };
4575 fid = &ll_i2info(inode)->lli_fid;
4576 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
4578 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
4579 fid, LDLM_IBITS, &policy, mode, lockh);
4584 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
4586 /* Already unlinked. Just update nlink and return success */
4587 if (rc == -ENOENT) {
4589 /* If it is striped directory, and there is bad stripe
4590 * Let's revalidate the dentry again, instead of returning
4592 if (ll_dir_striped(inode))
4595 /* This path cannot be hit for regular files unless in
4596 * case of obscure races, so no need to to validate
4598 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
4600 } else if (rc != 0) {
4601 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
4602 "%s: revalidate FID "DFID" error: rc = %d\n",
4603 ll_i2sbi(inode)->ll_fsname,
4604 PFID(ll_inode2fid(inode)), rc);
4610 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
4612 struct inode *inode = dentry->d_inode;
4613 struct obd_export *exp = ll_i2mdexp(inode);
4614 struct lookup_intent oit = {
4617 struct ptlrpc_request *req = NULL;
4618 struct md_op_data *op_data;
4622 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
4623 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
4625 /* Call getattr by fid, so do not provide name at all. */
4626 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
4627 LUSTRE_OPC_ANY, NULL);
4628 if (IS_ERR(op_data))
4629 RETURN(PTR_ERR(op_data));
4631 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
4632 ll_finish_md_op_data(op_data);
4634 rc = ll_inode_revalidate_fini(inode, rc);
4638 rc = ll_revalidate_it_finish(req, &oit, dentry);
4640 ll_intent_release(&oit);
4644 /* Unlinked? Unhash dentry, so it is not picked up later by
4645 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
4646 * here to preserve get_cwd functionality on 2.6.
4648 if (!dentry->d_inode->i_nlink) {
4649 spin_lock(&inode->i_lock);
4650 d_lustre_invalidate(dentry, 0);
4651 spin_unlock(&inode->i_lock);
4654 ll_lookup_finish_locks(&oit, dentry);
4656 ptlrpc_req_finished(req);
4661 static int ll_merge_md_attr(struct inode *inode)
4663 struct ll_inode_info *lli = ll_i2info(inode);
4664 struct cl_attr attr = { 0 };
4667 LASSERT(lli->lli_lsm_md != NULL);
4669 if (!lmv_dir_striped(lli->lli_lsm_md))
4672 down_read(&lli->lli_lsm_sem);
4673 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
4674 &attr, ll_md_blocking_ast);
4675 up_read(&lli->lli_lsm_sem);
4679 set_nlink(inode, attr.cat_nlink);
4680 inode->i_blocks = attr.cat_blocks;
4681 i_size_write(inode, attr.cat_size);
4683 ll_i2info(inode)->lli_atime = attr.cat_atime;
4684 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
4685 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
4690 int ll_getattr_dentry(struct dentry *de, struct kstat *stat)
4692 struct inode *inode = de->d_inode;
4693 struct ll_sb_info *sbi = ll_i2sbi(inode);
4694 struct ll_inode_info *lli = ll_i2info(inode);
4695 ktime_t kstart = ktime_get();
4698 rc = ll_inode_revalidate(de, IT_GETATTR);
4702 if (S_ISREG(inode->i_mode)) {
4705 rc = pcc_inode_getattr(inode, &cached);
4706 if (cached && rc < 0)
4709 /* In case of restore, the MDT has the right size and has
4710 * already send it back without granting the layout lock,
4711 * inode is up-to-date so glimpse is useless.
4712 * Also to glimpse we need the layout, in case of a running
4713 * restore the MDT holds the layout lock so the glimpse will
4714 * block up to the end of restore (getattr will block)
4716 if (!cached && !ll_file_test_flag(lli, LLIF_FILE_RESTORING)) {
4717 rc = ll_glimpse_size(inode);
4722 /* If object isn't regular a file then don't validate size. */
4723 if (ll_dir_striped(inode)) {
4724 rc = ll_merge_md_attr(inode);
4729 inode->i_atime.tv_sec = lli->lli_atime;
4730 inode->i_mtime.tv_sec = lli->lli_mtime;
4731 inode->i_ctime.tv_sec = lli->lli_ctime;
4734 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
4736 if (ll_need_32bit_api(sbi)) {
4737 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
4738 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
4739 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
4741 stat->ino = inode->i_ino;
4742 stat->dev = inode->i_sb->s_dev;
4743 stat->rdev = inode->i_rdev;
4746 stat->mode = inode->i_mode;
4747 stat->uid = inode->i_uid;
4748 stat->gid = inode->i_gid;
4749 stat->atime = inode->i_atime;
4750 stat->mtime = inode->i_mtime;
4751 stat->ctime = inode->i_ctime;
4752 /* stat->blksize is used to tell about preferred IO size */
4753 if (sbi->ll_stat_blksize)
4754 stat->blksize = sbi->ll_stat_blksize;
4755 else if (S_ISREG(inode->i_mode))
4756 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
4757 LL_MAX_BLKSIZE_BITS);
4759 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
4761 stat->nlink = inode->i_nlink;
4762 stat->size = i_size_read(inode);
4763 stat->blocks = inode->i_blocks;
4765 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
4766 ktime_us_delta(ktime_get(), kstart));
4771 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
4772 int ll_getattr(const struct path *path, struct kstat *stat,
4773 u32 request_mask, unsigned int flags)
4775 struct dentry *de = path->dentry;
4777 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
4780 return ll_getattr_dentry(de, stat);
4783 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4784 __u64 start, __u64 len)
4788 struct fiemap *fiemap;
4789 unsigned int extent_count = fieinfo->fi_extents_max;
4791 num_bytes = sizeof(*fiemap) + (extent_count *
4792 sizeof(struct fiemap_extent));
4793 OBD_ALLOC_LARGE(fiemap, num_bytes);
4798 fiemap->fm_flags = fieinfo->fi_flags;
4799 fiemap->fm_extent_count = fieinfo->fi_extents_max;
4800 fiemap->fm_start = start;
4801 fiemap->fm_length = len;
4802 if (extent_count > 0 &&
4803 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
4804 sizeof(struct fiemap_extent)) != 0)
4805 GOTO(out, rc = -EFAULT);
4807 rc = ll_do_fiemap(inode, fiemap, num_bytes);
4809 fieinfo->fi_flags = fiemap->fm_flags;
4810 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
4811 if (extent_count > 0 &&
4812 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
4813 fiemap->fm_mapped_extents *
4814 sizeof(struct fiemap_extent)) != 0)
4815 GOTO(out, rc = -EFAULT);
4817 OBD_FREE_LARGE(fiemap, num_bytes);
4821 struct posix_acl *ll_get_acl(struct inode *inode, int type)
4823 struct ll_inode_info *lli = ll_i2info(inode);
4824 struct posix_acl *acl = NULL;
4827 spin_lock(&lli->lli_lock);
4828 /* VFS' acl_permission_check->check_acl will release the refcount */
4829 acl = posix_acl_dup(lli->lli_posix_acl);
4830 spin_unlock(&lli->lli_lock);
4835 #ifdef HAVE_IOP_SET_ACL
4836 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
4837 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type)
4839 struct ll_sb_info *sbi = ll_i2sbi(inode);
4840 struct ptlrpc_request *req = NULL;
4841 const char *name = NULL;
4843 size_t value_size = 0;
4848 case ACL_TYPE_ACCESS:
4849 name = XATTR_NAME_POSIX_ACL_ACCESS;
4851 rc = posix_acl_update_mode(inode, &inode->i_mode, &acl);
4854 case ACL_TYPE_DEFAULT:
4855 name = XATTR_NAME_POSIX_ACL_DEFAULT;
4856 if (!S_ISDIR(inode->i_mode))
4857 rc = acl ? -EACCES : 0;
4868 value_size = posix_acl_xattr_size(acl->a_count);
4869 value = kmalloc(value_size, GFP_NOFS);
4871 GOTO(out, rc = -ENOMEM);
4873 rc = posix_acl_to_xattr(&init_user_ns, acl, value, value_size);
4875 GOTO(out_value, rc);
4878 rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode),
4879 value ? OBD_MD_FLXATTR : OBD_MD_FLXATTRRM,
4880 name, value, value_size, 0, 0, &req);
4882 ptlrpc_req_finished(req);
4887 forget_cached_acl(inode, type);
4889 set_cached_acl(inode, type, acl);
4892 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
4893 #endif /* HAVE_IOP_SET_ACL */
4895 int ll_inode_permission(struct inode *inode, int mask)
4898 struct ll_sb_info *sbi;
4899 struct root_squash_info *squash;
4900 struct cred *cred = NULL;
4901 const struct cred *old_cred = NULL;
4903 bool squash_id = false;
4904 ktime_t kstart = ktime_get();
4907 if (mask & MAY_NOT_BLOCK)
4910 /* as root inode are NOT getting validated in lookup operation,
4911 * need to do it before permission check. */
4913 if (inode == inode->i_sb->s_root->d_inode) {
4914 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
4919 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
4920 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
4922 /* squash fsuid/fsgid if needed */
4923 sbi = ll_i2sbi(inode);
4924 squash = &sbi->ll_squash;
4925 if (unlikely(squash->rsi_uid != 0 &&
4926 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
4927 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
4931 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
4932 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
4933 squash->rsi_uid, squash->rsi_gid);
4935 /* update current process's credentials
4936 * and FS capability */
4937 cred = prepare_creds();
4941 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
4942 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
4943 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
4944 if ((1 << cap) & CFS_CAP_FS_MASK)
4945 cap_lower(cred->cap_effective, cap);
4947 old_cred = override_creds(cred);
4950 rc = generic_permission(inode, mask);
4951 /* restore current process's credentials and FS capability */
4953 revert_creds(old_cred);
4958 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
4959 ktime_us_delta(ktime_get(), kstart));
4964 /* -o localflock - only provides locally consistent flock locks */
4965 struct file_operations ll_file_operations = {
4966 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
4967 # ifdef HAVE_SYNC_READ_WRITE
4968 .read = new_sync_read,
4969 .write = new_sync_write,
4971 .read_iter = ll_file_read_iter,
4972 .write_iter = ll_file_write_iter,
4973 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
4974 .read = ll_file_read,
4975 .aio_read = ll_file_aio_read,
4976 .write = ll_file_write,
4977 .aio_write = ll_file_aio_write,
4978 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
4979 .unlocked_ioctl = ll_file_ioctl,
4980 .open = ll_file_open,
4981 .release = ll_file_release,
4982 .mmap = ll_file_mmap,
4983 .llseek = ll_file_seek,
4984 .splice_read = ll_file_splice_read,
4989 struct file_operations ll_file_operations_flock = {
4990 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
4991 # ifdef HAVE_SYNC_READ_WRITE
4992 .read = new_sync_read,
4993 .write = new_sync_write,
4994 # endif /* HAVE_SYNC_READ_WRITE */
4995 .read_iter = ll_file_read_iter,
4996 .write_iter = ll_file_write_iter,
4997 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
4998 .read = ll_file_read,
4999 .aio_read = ll_file_aio_read,
5000 .write = ll_file_write,
5001 .aio_write = ll_file_aio_write,
5002 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5003 .unlocked_ioctl = ll_file_ioctl,
5004 .open = ll_file_open,
5005 .release = ll_file_release,
5006 .mmap = ll_file_mmap,
5007 .llseek = ll_file_seek,
5008 .splice_read = ll_file_splice_read,
5011 .flock = ll_file_flock,
5012 .lock = ll_file_flock
5015 /* These are for -o noflock - to return ENOSYS on flock calls */
5016 struct file_operations ll_file_operations_noflock = {
5017 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5018 # ifdef HAVE_SYNC_READ_WRITE
5019 .read = new_sync_read,
5020 .write = new_sync_write,
5021 # endif /* HAVE_SYNC_READ_WRITE */
5022 .read_iter = ll_file_read_iter,
5023 .write_iter = ll_file_write_iter,
5024 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5025 .read = ll_file_read,
5026 .aio_read = ll_file_aio_read,
5027 .write = ll_file_write,
5028 .aio_write = ll_file_aio_write,
5029 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5030 .unlocked_ioctl = ll_file_ioctl,
5031 .open = ll_file_open,
5032 .release = ll_file_release,
5033 .mmap = ll_file_mmap,
5034 .llseek = ll_file_seek,
5035 .splice_read = ll_file_splice_read,
5038 .flock = ll_file_noflock,
5039 .lock = ll_file_noflock
5042 struct inode_operations ll_file_inode_operations = {
5043 .setattr = ll_setattr,
5044 .getattr = ll_getattr,
5045 .permission = ll_inode_permission,
5046 #ifdef HAVE_IOP_XATTR
5047 .setxattr = ll_setxattr,
5048 .getxattr = ll_getxattr,
5049 .removexattr = ll_removexattr,
5051 .listxattr = ll_listxattr,
5052 .fiemap = ll_fiemap,
5053 .get_acl = ll_get_acl,
5054 #ifdef HAVE_IOP_SET_ACL
5055 .set_acl = ll_set_acl,
5059 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5061 struct ll_inode_info *lli = ll_i2info(inode);
5062 struct cl_object *obj = lli->lli_clob;
5071 env = cl_env_get(&refcheck);
5073 RETURN(PTR_ERR(env));
5075 rc = cl_conf_set(env, lli->lli_clob, conf);
5079 if (conf->coc_opc == OBJECT_CONF_SET) {
5080 struct ldlm_lock *lock = conf->coc_lock;
5081 struct cl_layout cl = {
5085 LASSERT(lock != NULL);
5086 LASSERT(ldlm_has_layout(lock));
5088 /* it can only be allowed to match after layout is
5089 * applied to inode otherwise false layout would be
5090 * seen. Applying layout shoud happen before dropping
5091 * the intent lock. */
5092 ldlm_lock_allow_match(lock);
5094 rc = cl_object_layout_get(env, obj, &cl);
5099 DFID": layout version change: %u -> %u\n",
5100 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5102 ll_layout_version_set(lli, cl.cl_layout_gen);
5106 cl_env_put(env, &refcheck);
5111 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5112 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5115 struct ll_sb_info *sbi = ll_i2sbi(inode);
5116 struct ptlrpc_request *req;
5123 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5124 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5125 lock->l_lvb_data, lock->l_lvb_len);
5127 if (lock->l_lvb_data != NULL)
5130 /* if layout lock was granted right away, the layout is returned
5131 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5132 * blocked and then granted via completion ast, we have to fetch
5133 * layout here. Please note that we can't use the LVB buffer in
5134 * completion AST because it doesn't have a large enough buffer */
5135 rc = ll_get_default_mdsize(sbi, &lmmsize);
5139 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5140 XATTR_NAME_LOV, lmmsize, &req);
5143 GOTO(out, rc = 0); /* empty layout */
5150 if (lmmsize == 0) /* empty layout */
5153 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5155 GOTO(out, rc = -EFAULT);
5157 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5158 if (lvbdata == NULL)
5159 GOTO(out, rc = -ENOMEM);
5161 memcpy(lvbdata, lmm, lmmsize);
5162 lock_res_and_lock(lock);
5163 if (unlikely(lock->l_lvb_data == NULL)) {
5164 lock->l_lvb_type = LVB_T_LAYOUT;
5165 lock->l_lvb_data = lvbdata;
5166 lock->l_lvb_len = lmmsize;
5169 unlock_res_and_lock(lock);
5172 OBD_FREE_LARGE(lvbdata, lmmsize);
5177 ptlrpc_req_finished(req);
5182 * Apply the layout to the inode. Layout lock is held and will be released
5185 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5186 struct inode *inode)
5188 struct ll_inode_info *lli = ll_i2info(inode);
5189 struct ll_sb_info *sbi = ll_i2sbi(inode);
5190 struct ldlm_lock *lock;
5191 struct cl_object_conf conf;
5194 bool wait_layout = false;
5197 LASSERT(lustre_handle_is_used(lockh));
5199 lock = ldlm_handle2lock(lockh);
5200 LASSERT(lock != NULL);
5201 LASSERT(ldlm_has_layout(lock));
5203 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5204 PFID(&lli->lli_fid), inode);
5206 /* in case this is a caching lock and reinstate with new inode */
5207 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5209 lock_res_and_lock(lock);
5210 lvb_ready = ldlm_is_lvb_ready(lock);
5211 unlock_res_and_lock(lock);
5213 /* checking lvb_ready is racy but this is okay. The worst case is
5214 * that multi processes may configure the file on the same time. */
5218 rc = ll_layout_fetch(inode, lock);
5222 /* for layout lock, lmm is stored in lock's lvb.
5223 * lvb_data is immutable if the lock is held so it's safe to access it
5226 * set layout to file. Unlikely this will fail as old layout was
5227 * surely eliminated */
5228 memset(&conf, 0, sizeof conf);
5229 conf.coc_opc = OBJECT_CONF_SET;
5230 conf.coc_inode = inode;
5231 conf.coc_lock = lock;
5232 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5233 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5234 rc = ll_layout_conf(inode, &conf);
5236 /* refresh layout failed, need to wait */
5237 wait_layout = rc == -EBUSY;
5240 LDLM_LOCK_PUT(lock);
5241 ldlm_lock_decref(lockh, mode);
5243 /* wait for IO to complete if it's still being used. */
5245 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5246 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5248 memset(&conf, 0, sizeof conf);
5249 conf.coc_opc = OBJECT_CONF_WAIT;
5250 conf.coc_inode = inode;
5251 rc = ll_layout_conf(inode, &conf);
5255 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5256 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5262 * Issue layout intent RPC to MDS.
5263 * \param inode [in] file inode
5264 * \param intent [in] layout intent
5266 * \retval 0 on success
5267 * \retval < 0 error code
5269 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5271 struct ll_inode_info *lli = ll_i2info(inode);
5272 struct ll_sb_info *sbi = ll_i2sbi(inode);
5273 struct md_op_data *op_data;
5274 struct lookup_intent it;
5275 struct ptlrpc_request *req;
5279 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5280 0, 0, LUSTRE_OPC_ANY, NULL);
5281 if (IS_ERR(op_data))
5282 RETURN(PTR_ERR(op_data));
5284 op_data->op_data = intent;
5285 op_data->op_data_size = sizeof(*intent);
5287 memset(&it, 0, sizeof(it));
5288 it.it_op = IT_LAYOUT;
5289 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5290 intent->li_opc == LAYOUT_INTENT_TRUNC)
5291 it.it_flags = FMODE_WRITE;
5293 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5294 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5296 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5297 &ll_md_blocking_ast, 0);
5298 if (it.it_request != NULL)
5299 ptlrpc_req_finished(it.it_request);
5300 it.it_request = NULL;
5302 ll_finish_md_op_data(op_data);
5304 /* set lock data in case this is a new lock */
5306 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5308 ll_intent_drop_lock(&it);
5314 * This function checks if there exists a LAYOUT lock on the client side,
5315 * or enqueues it if it doesn't have one in cache.
5317 * This function will not hold layout lock so it may be revoked any time after
5318 * this function returns. Any operations depend on layout should be redone
5321 * This function should be called before lov_io_init() to get an uptodate
5322 * layout version, the caller should save the version number and after IO
5323 * is finished, this function should be called again to verify that layout
5324 * is not changed during IO time.
5326 int ll_layout_refresh(struct inode *inode, __u32 *gen)
5328 struct ll_inode_info *lli = ll_i2info(inode);
5329 struct ll_sb_info *sbi = ll_i2sbi(inode);
5330 struct lustre_handle lockh;
5331 struct layout_intent intent = {
5332 .li_opc = LAYOUT_INTENT_ACCESS,
5334 enum ldlm_mode mode;
5338 *gen = ll_layout_version_get(lli);
5339 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
5343 LASSERT(fid_is_sane(ll_inode2fid(inode)));
5344 LASSERT(S_ISREG(inode->i_mode));
5346 /* take layout lock mutex to enqueue layout lock exclusively. */
5347 mutex_lock(&lli->lli_layout_mutex);
5350 /* mostly layout lock is caching on the local side, so try to
5351 * match it before grabbing layout lock mutex. */
5352 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
5353 LCK_CR | LCK_CW | LCK_PR |
5355 if (mode != 0) { /* hit cached lock */
5356 rc = ll_layout_lock_set(&lockh, mode, inode);
5362 rc = ll_layout_intent(inode, &intent);
5368 *gen = ll_layout_version_get(lli);
5369 mutex_unlock(&lli->lli_layout_mutex);
5375 * Issue layout intent RPC indicating where in a file an IO is about to write.
5377 * \param[in] inode file inode.
5378 * \param[in] ext write range with start offset of fille in bytes where
5379 * an IO is about to write, and exclusive end offset in
5382 * \retval 0 on success
5383 * \retval < 0 error code
5385 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
5386 struct lu_extent *ext)
5388 struct layout_intent intent = {
5390 .li_extent.e_start = ext->e_start,
5391 .li_extent.e_end = ext->e_end,
5396 rc = ll_layout_intent(inode, &intent);
5402 * This function send a restore request to the MDT
5404 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
5406 struct hsm_user_request *hur;
5410 len = sizeof(struct hsm_user_request) +
5411 sizeof(struct hsm_user_item);
5412 OBD_ALLOC(hur, len);
5416 hur->hur_request.hr_action = HUA_RESTORE;
5417 hur->hur_request.hr_archive_id = 0;
5418 hur->hur_request.hr_flags = 0;
5419 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
5420 sizeof(hur->hur_user_item[0].hui_fid));
5421 hur->hur_user_item[0].hui_extent.offset = offset;
5422 hur->hur_user_item[0].hui_extent.length = length;
5423 hur->hur_request.hr_itemcount = 1;
5424 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,