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>
46 #include <linux/falloc.h>
48 #include <uapi/linux/lustre/lustre_ioctl.h>
49 #include <uapi/linux/llcrypt.h>
50 #include <lustre_swab.h>
52 #include "cl_object.h"
53 #include "llite_internal.h"
54 #include "vvp_internal.h"
57 struct inode *sp_inode;
62 __u64 pa_data_version;
68 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
70 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
73 static struct ll_file_data *ll_file_data_get(void)
75 struct ll_file_data *fd;
77 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
81 fd->fd_write_failed = false;
82 pcc_file_init(&fd->fd_pcc_file);
87 static void ll_file_data_put(struct ll_file_data *fd)
90 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
94 * Packs all the attributes into @op_data for the CLOSE rpc.
96 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
97 struct obd_client_handle *och)
101 ll_prep_md_op_data(op_data, inode, NULL, NULL,
102 0, 0, LUSTRE_OPC_ANY, NULL);
104 op_data->op_attr.ia_mode = inode->i_mode;
105 op_data->op_attr.ia_atime = inode->i_atime;
106 op_data->op_attr.ia_mtime = inode->i_mtime;
107 op_data->op_attr.ia_ctime = inode->i_ctime;
108 op_data->op_attr.ia_size = i_size_read(inode);
109 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
110 ATTR_MTIME | ATTR_MTIME_SET |
112 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
113 op_data->op_attr_blocks = inode->i_blocks;
114 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
115 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
116 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
117 op_data->op_open_handle = och->och_open_handle;
119 if (och->och_flags & FMODE_WRITE &&
120 ll_file_test_and_clear_flag(ll_i2info(inode), LLIF_DATA_MODIFIED))
121 /* For HSM: if inode data has been modified, pack it so that
122 * MDT can set data dirty flag in the archive. */
123 op_data->op_bias |= MDS_DATA_MODIFIED;
129 * Perform a close, possibly with a bias.
130 * The meaning of "data" depends on the value of "bias".
132 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
133 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
136 static int ll_close_inode_openhandle(struct inode *inode,
137 struct obd_client_handle *och,
138 enum mds_op_bias bias, void *data)
140 struct obd_export *md_exp = ll_i2mdexp(inode);
141 const struct ll_inode_info *lli = ll_i2info(inode);
142 struct md_op_data *op_data;
143 struct ptlrpc_request *req = NULL;
147 if (class_exp2obd(md_exp) == NULL) {
148 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
149 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
153 OBD_ALLOC_PTR(op_data);
154 /* We leak openhandle and request here on error, but not much to be
155 * done in OOM case since app won't retry close on error either. */
157 GOTO(out, rc = -ENOMEM);
159 ll_prepare_close(inode, op_data, och);
161 case MDS_CLOSE_LAYOUT_MERGE:
162 /* merge blocks from the victim inode */
163 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
164 op_data->op_attr.ia_valid |= ATTR_SIZE;
165 op_data->op_xvalid |= OP_XVALID_BLOCKS;
167 case MDS_CLOSE_LAYOUT_SPLIT:
168 case MDS_CLOSE_LAYOUT_SWAP: {
169 struct split_param *sp = data;
171 LASSERT(data != NULL);
172 op_data->op_bias |= bias;
173 op_data->op_data_version = 0;
174 op_data->op_lease_handle = och->och_lease_handle;
175 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
176 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
177 op_data->op_mirror_id = sp->sp_mirror_id;
179 op_data->op_fid2 = *ll_inode2fid(data);
184 case MDS_CLOSE_RESYNC_DONE: {
185 struct ll_ioc_lease *ioc = data;
187 LASSERT(data != NULL);
188 op_data->op_attr_blocks +=
189 ioc->lil_count * op_data->op_attr_blocks;
190 op_data->op_attr.ia_valid |= ATTR_SIZE;
191 op_data->op_xvalid |= OP_XVALID_BLOCKS;
192 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
194 op_data->op_lease_handle = och->och_lease_handle;
195 op_data->op_data = &ioc->lil_ids[0];
196 op_data->op_data_size =
197 ioc->lil_count * sizeof(ioc->lil_ids[0]);
201 case MDS_PCC_ATTACH: {
202 struct pcc_param *param = data;
204 LASSERT(data != NULL);
205 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
206 op_data->op_archive_id = param->pa_archive_id;
207 op_data->op_data_version = param->pa_data_version;
208 op_data->op_lease_handle = och->och_lease_handle;
212 case MDS_HSM_RELEASE:
213 LASSERT(data != NULL);
214 op_data->op_bias |= MDS_HSM_RELEASE;
215 op_data->op_data_version = *(__u64 *)data;
216 op_data->op_lease_handle = och->och_lease_handle;
217 op_data->op_attr.ia_valid |= ATTR_SIZE;
218 op_data->op_xvalid |= OP_XVALID_BLOCKS;
222 LASSERT(data == NULL);
226 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
227 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
228 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
229 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
231 rc = md_close(md_exp, op_data, och->och_mod, &req);
232 if (rc != 0 && rc != -EINTR)
233 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
234 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
236 if (rc == 0 && op_data->op_bias & bias) {
237 struct mdt_body *body;
239 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
240 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
243 if (bias & MDS_PCC_ATTACH) {
244 struct pcc_param *param = data;
246 param->pa_layout_gen = body->mbo_layout_gen;
250 ll_finish_md_op_data(op_data);
254 md_clear_open_replay_data(md_exp, och);
255 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
258 ptlrpc_req_finished(req); /* This is close request */
262 int ll_md_real_close(struct inode *inode, fmode_t fmode)
264 struct ll_inode_info *lli = ll_i2info(inode);
265 struct obd_client_handle **och_p;
266 struct obd_client_handle *och;
271 if (fmode & FMODE_WRITE) {
272 och_p = &lli->lli_mds_write_och;
273 och_usecount = &lli->lli_open_fd_write_count;
274 } else if (fmode & FMODE_EXEC) {
275 och_p = &lli->lli_mds_exec_och;
276 och_usecount = &lli->lli_open_fd_exec_count;
278 LASSERT(fmode & FMODE_READ);
279 och_p = &lli->lli_mds_read_och;
280 och_usecount = &lli->lli_open_fd_read_count;
283 mutex_lock(&lli->lli_och_mutex);
284 if (*och_usecount > 0) {
285 /* There are still users of this handle, so skip
287 mutex_unlock(&lli->lli_och_mutex);
293 mutex_unlock(&lli->lli_och_mutex);
296 /* There might be a race and this handle may already
298 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
304 static int ll_md_close(struct inode *inode, struct file *file)
306 union ldlm_policy_data policy = {
307 .l_inodebits = { MDS_INODELOCK_OPEN },
309 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
310 struct ll_file_data *fd = file->private_data;
311 struct ll_inode_info *lli = ll_i2info(inode);
312 struct lustre_handle lockh;
313 enum ldlm_mode lockmode;
317 /* clear group lock, if present */
318 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
319 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
321 if (fd->fd_lease_och != NULL) {
324 /* Usually the lease is not released when the
325 * application crashed, we need to release here. */
326 rc = ll_lease_close(fd->fd_lease_och, inode, &lease_broken);
327 CDEBUG(rc ? D_ERROR : D_INODE, "Clean up lease "DFID" %d/%d\n",
328 PFID(&lli->lli_fid), rc, lease_broken);
330 fd->fd_lease_och = NULL;
333 if (fd->fd_och != NULL) {
334 rc = ll_close_inode_openhandle(inode, fd->fd_och, 0, NULL);
339 /* Let's see if we have good enough OPEN lock on the file and if
340 we can skip talking to MDS */
341 mutex_lock(&lli->lli_och_mutex);
342 if (fd->fd_omode & FMODE_WRITE) {
344 LASSERT(lli->lli_open_fd_write_count);
345 lli->lli_open_fd_write_count--;
346 } else if (fd->fd_omode & FMODE_EXEC) {
348 LASSERT(lli->lli_open_fd_exec_count);
349 lli->lli_open_fd_exec_count--;
352 LASSERT(lli->lli_open_fd_read_count);
353 lli->lli_open_fd_read_count--;
355 mutex_unlock(&lli->lli_och_mutex);
357 /* LU-4398: do not cache write open lock if the file has exec bit */
358 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
359 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
360 LDLM_IBITS, &policy, lockmode, &lockh))
361 rc = ll_md_real_close(inode, fd->fd_omode);
364 file->private_data = NULL;
365 ll_file_data_put(fd);
370 /* While this returns an error code, fput() the caller does not, so we need
371 * to make every effort to clean up all of our state here. Also, applications
372 * rarely check close errors and even if an error is returned they will not
373 * re-try the close call.
375 int ll_file_release(struct inode *inode, struct file *file)
377 struct ll_file_data *fd;
378 struct ll_sb_info *sbi = ll_i2sbi(inode);
379 struct ll_inode_info *lli = ll_i2info(inode);
380 ktime_t kstart = ktime_get();
385 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
386 PFID(ll_inode2fid(inode)), inode);
388 fd = file->private_data;
391 /* The last ref on @file, maybe not the the owner pid of statahead,
392 * because parent and child process can share the same file handle. */
393 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
394 ll_deauthorize_statahead(inode, fd);
396 if (inode->i_sb->s_root == file_dentry(file)) {
397 file->private_data = NULL;
398 ll_file_data_put(fd);
402 pcc_file_release(inode, file);
404 if (!S_ISDIR(inode->i_mode)) {
405 if (lli->lli_clob != NULL)
406 lov_read_and_clear_async_rc(lli->lli_clob);
407 lli->lli_async_rc = 0;
410 rc = ll_md_close(inode, file);
412 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
413 libcfs_debug_dumplog();
416 if (!rc && inode->i_sb->s_root != file_dentry(file))
417 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
418 ktime_us_delta(ktime_get(), kstart));
422 static inline int ll_dom_readpage(void *data, struct page *page)
424 struct niobuf_local *lnb = data;
428 struct inode *inode = page2inode(page);
430 kaddr = kmap_atomic(page);
431 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
432 if (lnb->lnb_len < PAGE_SIZE)
433 memset(kaddr + lnb->lnb_len, 0,
434 PAGE_SIZE - lnb->lnb_len);
435 flush_dcache_page(page);
436 SetPageUptodate(page);
437 kunmap_atomic(kaddr);
439 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
440 if (!llcrypt_has_encryption_key(inode))
441 CDEBUG(D_SEC, "no enc key for "DFID"\n",
442 PFID(ll_inode2fid(inode)));
444 unsigned int offs = 0;
446 while (offs < PAGE_SIZE) {
447 /* decrypt only if page is not empty */
448 if (memcmp(page_address(page) + offs,
449 page_address(ZERO_PAGE(0)),
450 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
453 rc = llcrypt_decrypt_pagecache_blocks(page,
454 LUSTRE_ENCRYPTION_UNIT_SIZE,
459 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
468 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
472 struct ll_inode_info *lli = ll_i2info(inode);
473 struct cl_object *obj = lli->lli_clob;
474 struct address_space *mapping = inode->i_mapping;
476 struct niobuf_remote *rnb;
477 struct mdt_body *body;
479 unsigned long index, start;
480 struct niobuf_local lnb;
489 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
493 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
494 if (rnb == NULL || rnb->rnb_len == 0)
497 /* LU-11595: Server may return whole file and that is OK always or
498 * it may return just file tail and its offset must be aligned with
499 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
500 * smaller then offset may be not aligned and that data is just ignored.
502 if (rnb->rnb_offset & ~PAGE_MASK)
505 /* Server returns whole file or just file tail if it fills in reply
506 * buffer, in both cases total size should be equal to the file size.
508 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
509 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
510 !(inode && IS_ENCRYPTED(inode))) {
511 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
512 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
513 rnb->rnb_len, body->mbo_dom_size);
517 env = cl_env_get(&refcheck);
520 io = vvp_env_thread_io(env);
522 io->ci_ignore_layout = 1;
523 rc = cl_io_init(env, io, CIT_MISC, obj);
527 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
528 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
530 data = (char *)rnb + sizeof(*rnb);
532 lnb.lnb_file_offset = rnb->rnb_offset;
533 start = lnb.lnb_file_offset >> PAGE_SHIFT;
535 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
536 lnb.lnb_page_offset = 0;
538 struct cl_page *page;
540 lnb.lnb_data = data + (index << PAGE_SHIFT);
541 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
542 if (lnb.lnb_len > PAGE_SIZE)
543 lnb.lnb_len = PAGE_SIZE;
545 vmpage = read_cache_page(mapping, index + start,
546 ll_dom_readpage, &lnb);
547 if (IS_ERR(vmpage)) {
548 CWARN("%s: cannot fill page %lu for "DFID
549 " with data: rc = %li\n",
550 ll_i2sbi(inode)->ll_fsname, index + start,
551 PFID(lu_object_fid(&obj->co_lu)),
556 if (vmpage->mapping == NULL) {
559 /* page was truncated */
562 /* attach VM page to CL page cache */
563 page = cl_page_find(env, obj, vmpage->index, vmpage,
566 ClearPageUptodate(vmpage);
571 cl_page_export(env, page, 1);
572 cl_page_put(env, page);
576 } while (rnb->rnb_len > (index << PAGE_SHIFT));
580 cl_env_put(env, &refcheck);
585 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
586 struct lookup_intent *itp)
588 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
589 struct dentry *parent = de->d_parent;
592 struct md_op_data *op_data;
593 struct ptlrpc_request *req = NULL;
597 LASSERT(parent != NULL);
598 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
600 /* if server supports open-by-fid, or file name is invalid, don't pack
601 * name in open request */
602 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
603 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
605 len = de->d_name.len;
606 name = kmalloc(len + 1, GFP_NOFS);
611 spin_lock(&de->d_lock);
612 if (len != de->d_name.len) {
613 spin_unlock(&de->d_lock);
617 memcpy(name, de->d_name.name, len);
619 spin_unlock(&de->d_lock);
621 if (!lu_name_is_valid_2(name, len)) {
627 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
628 name, len, 0, LUSTRE_OPC_ANY, NULL);
629 if (IS_ERR(op_data)) {
631 RETURN(PTR_ERR(op_data));
633 op_data->op_data = lmm;
634 op_data->op_data_size = lmmsize;
636 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
637 &ll_md_blocking_ast, 0);
639 ll_finish_md_op_data(op_data);
641 /* reason for keep own exit path - don`t flood log
642 * with messages with -ESTALE errors.
644 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
645 it_open_error(DISP_OPEN_OPEN, itp))
647 ll_release_openhandle(de, itp);
651 if (it_disposition(itp, DISP_LOOKUP_NEG))
652 GOTO(out, rc = -ENOENT);
654 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
655 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
656 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
660 rc = ll_prep_inode(&de->d_inode, req, NULL, itp);
662 if (!rc && itp->it_lock_mode) {
665 /* If we got a lock back and it has a LOOKUP bit set,
666 * make sure the dentry is marked as valid so we can find it.
667 * We don't need to care about actual hashing since other bits
668 * of kernel will deal with that later.
670 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
671 if (bits & MDS_INODELOCK_LOOKUP)
672 d_lustre_revalidate(de);
673 /* if DoM bit returned along with LAYOUT bit then there
674 * can be read-on-open data returned.
676 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
677 ll_dom_finish_open(de->d_inode, req);
681 ptlrpc_req_finished(req);
682 ll_intent_drop_lock(itp);
684 /* We did open by fid, but by the time we got to the server,
685 * the object disappeared. If this is a create, we cannot really
686 * tell the userspace that the file it was trying to create
687 * does not exist. Instead let's return -ESTALE, and the VFS will
688 * retry the create with LOOKUP_REVAL that we are going to catch
689 * in ll_revalidate_dentry() and use lookup then.
691 if (rc == -ENOENT && itp->it_op & IT_CREAT)
697 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
698 struct obd_client_handle *och)
700 struct mdt_body *body;
702 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
703 och->och_open_handle = body->mbo_open_handle;
704 och->och_fid = body->mbo_fid1;
705 och->och_lease_handle.cookie = it->it_lock_handle;
706 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
707 och->och_flags = it->it_flags;
709 return md_set_open_replay_data(md_exp, och, it);
712 static int ll_local_open(struct file *file, struct lookup_intent *it,
713 struct ll_file_data *fd, struct obd_client_handle *och)
715 struct inode *inode = file_inode(file);
718 LASSERT(!file->private_data);
725 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
730 file->private_data = fd;
731 ll_readahead_init(inode, &fd->fd_ras);
732 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
734 /* ll_cl_context initialize */
735 rwlock_init(&fd->fd_lock);
736 INIT_LIST_HEAD(&fd->fd_lccs);
741 /* Open a file, and (for the very first open) create objects on the OSTs at
742 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
743 * creation or open until ll_lov_setstripe() ioctl is called.
745 * If we already have the stripe MD locally then we don't request it in
746 * md_open(), by passing a lmm_size = 0.
748 * It is up to the application to ensure no other processes open this file
749 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
750 * used. We might be able to avoid races of that sort by getting lli_open_sem
751 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
752 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
754 int ll_file_open(struct inode *inode, struct file *file)
756 struct ll_inode_info *lli = ll_i2info(inode);
757 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
758 .it_flags = file->f_flags };
759 struct obd_client_handle **och_p = NULL;
760 __u64 *och_usecount = NULL;
761 struct ll_file_data *fd;
762 ktime_t kstart = ktime_get();
766 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
767 PFID(ll_inode2fid(inode)), inode, file->f_flags);
769 it = file->private_data; /* XXX: compat macro */
770 file->private_data = NULL; /* prevent ll_local_open assertion */
772 if (S_ISREG(inode->i_mode)) {
773 rc = llcrypt_file_open(inode, file);
775 GOTO(out_nofiledata, rc);
778 fd = ll_file_data_get();
780 GOTO(out_nofiledata, rc = -ENOMEM);
783 if (S_ISDIR(inode->i_mode))
784 ll_authorize_statahead(inode, fd);
786 if (inode->i_sb->s_root == file_dentry(file)) {
787 file->private_data = fd;
791 if (!it || !it->it_disposition) {
792 /* Convert f_flags into access mode. We cannot use file->f_mode,
793 * because everything but O_ACCMODE mask was stripped from
795 if ((oit.it_flags + 1) & O_ACCMODE)
797 if (file->f_flags & O_TRUNC)
798 oit.it_flags |= FMODE_WRITE;
800 /* kernel only call f_op->open in dentry_open. filp_open calls
801 * dentry_open after call to open_namei that checks permissions.
802 * Only nfsd_open call dentry_open directly without checking
803 * permissions and because of that this code below is safe.
805 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
806 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
808 /* We do not want O_EXCL here, presumably we opened the file
809 * already? XXX - NFS implications? */
810 oit.it_flags &= ~O_EXCL;
812 /* bug20584, if "it_flags" contains O_CREAT, the file will be
813 * created if necessary, then "IT_CREAT" should be set to keep
814 * consistent with it */
815 if (oit.it_flags & O_CREAT)
816 oit.it_op |= IT_CREAT;
822 /* Let's see if we have file open on MDS already. */
823 if (it->it_flags & FMODE_WRITE) {
824 och_p = &lli->lli_mds_write_och;
825 och_usecount = &lli->lli_open_fd_write_count;
826 } else if (it->it_flags & FMODE_EXEC) {
827 och_p = &lli->lli_mds_exec_och;
828 och_usecount = &lli->lli_open_fd_exec_count;
830 och_p = &lli->lli_mds_read_och;
831 och_usecount = &lli->lli_open_fd_read_count;
834 mutex_lock(&lli->lli_och_mutex);
835 if (*och_p) { /* Open handle is present */
836 if (it_disposition(it, DISP_OPEN_OPEN)) {
837 /* Well, there's extra open request that we do not need,
838 * let's close it somehow. This will decref request. */
839 rc = it_open_error(DISP_OPEN_OPEN, it);
841 mutex_unlock(&lli->lli_och_mutex);
842 GOTO(out_openerr, rc);
845 ll_release_openhandle(file_dentry(file), it);
849 rc = ll_local_open(file, it, fd, NULL);
852 mutex_unlock(&lli->lli_och_mutex);
853 GOTO(out_openerr, rc);
856 LASSERT(*och_usecount == 0);
857 if (!it->it_disposition) {
858 struct dentry *dentry = file_dentry(file);
859 struct ll_dentry_data *ldd;
861 /* We cannot just request lock handle now, new ELC code
862 * means that one of other OPEN locks for this file
863 * could be cancelled, and since blocking ast handler
864 * would attempt to grab och_mutex as well, that would
865 * result in a deadlock
867 mutex_unlock(&lli->lli_och_mutex);
869 * Normally called under two situations:
871 * 2. A race/condition on MDS resulting in no open
872 * handle to be returned from LOOKUP|OPEN request,
873 * for example if the target entry was a symlink.
875 * Only fetch MDS_OPEN_LOCK if this is in NFS path,
876 * marked by a bit set in ll_iget_for_nfs. Clear the
877 * bit so that it's not confusing later callers.
879 * NB; when ldd is NULL, it must have come via normal
880 * lookup path only, since ll_iget_for_nfs always calls
883 ldd = ll_d2d(dentry);
884 if (ldd && ldd->lld_nfs_dentry) {
885 ldd->lld_nfs_dentry = 0;
886 if (!filename_is_volatile(dentry->d_name.name,
889 it->it_flags |= MDS_OPEN_LOCK;
893 * Always specify MDS_OPEN_BY_FID because we don't want
894 * to get file with different fid.
896 it->it_flags |= MDS_OPEN_BY_FID;
897 rc = ll_intent_file_open(dentry, NULL, 0, it);
899 GOTO(out_openerr, rc);
903 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
905 GOTO(out_och_free, rc = -ENOMEM);
909 /* md_intent_lock() didn't get a request ref if there was an
910 * open error, so don't do cleanup on the request here
912 /* XXX (green): Should not we bail out on any error here, not
913 * just open error? */
914 rc = it_open_error(DISP_OPEN_OPEN, it);
916 GOTO(out_och_free, rc);
918 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
919 "inode %p: disposition %x, status %d\n", inode,
920 it_disposition(it, ~0), it->it_status);
922 rc = ll_local_open(file, it, fd, *och_p);
924 GOTO(out_och_free, rc);
927 rc = pcc_file_open(inode, file);
929 GOTO(out_och_free, rc);
931 mutex_unlock(&lli->lli_och_mutex);
933 /* lockless for direct IO so that it can do IO in parallel */
934 if (file->f_flags & O_DIRECT)
935 fd->fd_flags |= LL_FILE_LOCKLESS_IO;
938 /* Must do this outside lli_och_mutex lock to prevent deadlock where
939 different kind of OPEN lock for this same inode gets cancelled
940 by ldlm_cancel_lru */
941 if (!S_ISREG(inode->i_mode))
942 GOTO(out_och_free, rc);
943 cl_lov_delay_create_clear(&file->f_flags);
944 GOTO(out_och_free, rc);
948 if (och_p && *och_p) {
949 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
950 *och_p = NULL; /* OBD_FREE writes some magic there */
953 mutex_unlock(&lli->lli_och_mutex);
956 if (lli->lli_opendir_key == fd)
957 ll_deauthorize_statahead(inode, fd);
960 ll_file_data_put(fd);
962 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
963 ktime_us_delta(ktime_get(), kstart));
967 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
968 ptlrpc_req_finished(it->it_request);
969 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
975 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
976 struct ldlm_lock_desc *desc, void *data, int flag)
979 struct lustre_handle lockh;
983 case LDLM_CB_BLOCKING:
984 ldlm_lock2handle(lock, &lockh);
985 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
987 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
991 case LDLM_CB_CANCELING:
999 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1000 * and save it as fd->fd_och so as to force client to reopen the file even
1001 * if it has an open lock in cache already.
1003 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1004 struct lustre_handle *old_open_handle)
1006 struct ll_inode_info *lli = ll_i2info(inode);
1007 struct ll_file_data *fd = file->private_data;
1008 struct obd_client_handle **och_p;
1009 __u64 *och_usecount;
1013 /* Get the openhandle of the file */
1014 mutex_lock(&lli->lli_och_mutex);
1015 if (fd->fd_lease_och != NULL)
1016 GOTO(out_unlock, rc = -EBUSY);
1018 if (fd->fd_och == NULL) {
1019 if (file->f_mode & FMODE_WRITE) {
1020 LASSERT(lli->lli_mds_write_och != NULL);
1021 och_p = &lli->lli_mds_write_och;
1022 och_usecount = &lli->lli_open_fd_write_count;
1024 LASSERT(lli->lli_mds_read_och != NULL);
1025 och_p = &lli->lli_mds_read_och;
1026 och_usecount = &lli->lli_open_fd_read_count;
1029 if (*och_usecount > 1)
1030 GOTO(out_unlock, rc = -EBUSY);
1032 fd->fd_och = *och_p;
1037 *old_open_handle = fd->fd_och->och_open_handle;
1041 mutex_unlock(&lli->lli_och_mutex);
1046 * Release ownership on lli_mds_*_och when putting back a file lease.
1048 static int ll_lease_och_release(struct inode *inode, struct file *file)
1050 struct ll_inode_info *lli = ll_i2info(inode);
1051 struct ll_file_data *fd = file->private_data;
1052 struct obd_client_handle **och_p;
1053 struct obd_client_handle *old_och = NULL;
1054 __u64 *och_usecount;
1058 mutex_lock(&lli->lli_och_mutex);
1059 if (file->f_mode & FMODE_WRITE) {
1060 och_p = &lli->lli_mds_write_och;
1061 och_usecount = &lli->lli_open_fd_write_count;
1063 och_p = &lli->lli_mds_read_och;
1064 och_usecount = &lli->lli_open_fd_read_count;
1067 /* The file may have been open by another process (broken lease) so
1068 * *och_p is not NULL. In this case we should simply increase usecount
1071 if (*och_p != NULL) {
1072 old_och = fd->fd_och;
1075 *och_p = fd->fd_och;
1079 mutex_unlock(&lli->lli_och_mutex);
1081 if (old_och != NULL)
1082 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1088 * Acquire a lease and open the file.
1090 static struct obd_client_handle *
1091 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1094 struct lookup_intent it = { .it_op = IT_OPEN };
1095 struct ll_sb_info *sbi = ll_i2sbi(inode);
1096 struct md_op_data *op_data;
1097 struct ptlrpc_request *req = NULL;
1098 struct lustre_handle old_open_handle = { 0 };
1099 struct obd_client_handle *och = NULL;
1104 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1105 RETURN(ERR_PTR(-EINVAL));
1108 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1109 RETURN(ERR_PTR(-EPERM));
1111 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1113 RETURN(ERR_PTR(rc));
1118 RETURN(ERR_PTR(-ENOMEM));
1120 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1121 LUSTRE_OPC_ANY, NULL);
1122 if (IS_ERR(op_data))
1123 GOTO(out, rc = PTR_ERR(op_data));
1125 /* To tell the MDT this openhandle is from the same owner */
1126 op_data->op_open_handle = old_open_handle;
1128 it.it_flags = fmode | open_flags;
1129 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1130 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1131 &ll_md_blocking_lease_ast,
1132 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1133 * it can be cancelled which may mislead applications that the lease is
1135 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1136 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1137 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1138 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1139 ll_finish_md_op_data(op_data);
1140 ptlrpc_req_finished(req);
1142 GOTO(out_release_it, rc);
1144 if (it_disposition(&it, DISP_LOOKUP_NEG))
1145 GOTO(out_release_it, rc = -ENOENT);
1147 rc = it_open_error(DISP_OPEN_OPEN, &it);
1149 GOTO(out_release_it, rc);
1151 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1152 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1154 GOTO(out_release_it, rc);
1156 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1157 GOTO(out_close, rc = -EOPNOTSUPP);
1159 /* already get lease, handle lease lock */
1160 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1161 if (!it.it_lock_mode ||
1162 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1163 /* open lock must return for lease */
1164 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1165 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1167 GOTO(out_close, rc = -EPROTO);
1170 ll_intent_release(&it);
1174 /* Cancel open lock */
1175 if (it.it_lock_mode != 0) {
1176 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1178 it.it_lock_mode = 0;
1179 och->och_lease_handle.cookie = 0ULL;
1181 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1183 CERROR("%s: error closing file "DFID": %d\n",
1184 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1185 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1187 ll_intent_release(&it);
1191 RETURN(ERR_PTR(rc));
1195 * Check whether a layout swap can be done between two inodes.
1197 * \param[in] inode1 First inode to check
1198 * \param[in] inode2 Second inode to check
1200 * \retval 0 on success, layout swap can be performed between both inodes
1201 * \retval negative error code if requirements are not met
1203 static int ll_check_swap_layouts_validity(struct inode *inode1,
1204 struct inode *inode2)
1206 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1209 if (inode_permission(inode1, MAY_WRITE) ||
1210 inode_permission(inode2, MAY_WRITE))
1213 if (inode1->i_sb != inode2->i_sb)
1219 static int ll_swap_layouts_close(struct obd_client_handle *och,
1220 struct inode *inode, struct inode *inode2)
1222 const struct lu_fid *fid1 = ll_inode2fid(inode);
1223 const struct lu_fid *fid2;
1227 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1228 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1230 rc = ll_check_swap_layouts_validity(inode, inode2);
1232 GOTO(out_free_och, rc);
1234 /* We now know that inode2 is a lustre inode */
1235 fid2 = ll_inode2fid(inode2);
1237 rc = lu_fid_cmp(fid1, fid2);
1239 GOTO(out_free_och, rc = -EINVAL);
1241 /* Close the file and {swap,merge} layouts between inode & inode2.
1242 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1243 * because we still need it to pack l_remote_handle to MDT. */
1244 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1247 och = NULL; /* freed in ll_close_inode_openhandle() */
1257 * Release lease and close the file.
1258 * It will check if the lease has ever broken.
1260 static int ll_lease_close_intent(struct obd_client_handle *och,
1261 struct inode *inode,
1262 bool *lease_broken, enum mds_op_bias bias,
1265 struct ldlm_lock *lock;
1266 bool cancelled = true;
1270 lock = ldlm_handle2lock(&och->och_lease_handle);
1272 lock_res_and_lock(lock);
1273 cancelled = ldlm_is_cancel(lock);
1274 unlock_res_and_lock(lock);
1275 LDLM_LOCK_PUT(lock);
1278 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1279 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1281 if (lease_broken != NULL)
1282 *lease_broken = cancelled;
1284 if (!cancelled && !bias)
1285 ldlm_cli_cancel(&och->och_lease_handle, 0);
1287 if (cancelled) { /* no need to excute intent */
1292 rc = ll_close_inode_openhandle(inode, och, bias, data);
1296 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1299 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1303 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1305 static int ll_lease_file_resync(struct obd_client_handle *och,
1306 struct inode *inode, unsigned long arg)
1308 struct ll_sb_info *sbi = ll_i2sbi(inode);
1309 struct md_op_data *op_data;
1310 struct ll_ioc_lease_id ioc;
1311 __u64 data_version_unused;
1315 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1316 LUSTRE_OPC_ANY, NULL);
1317 if (IS_ERR(op_data))
1318 RETURN(PTR_ERR(op_data));
1320 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1324 /* before starting file resync, it's necessary to clean up page cache
1325 * in client memory, otherwise once the layout version is increased,
1326 * writing back cached data will be denied the OSTs. */
1327 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1331 op_data->op_lease_handle = och->och_lease_handle;
1332 op_data->op_mirror_id = ioc.lil_mirror_id;
1333 rc = md_file_resync(sbi->ll_md_exp, op_data);
1339 ll_finish_md_op_data(op_data);
1343 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1345 struct ll_inode_info *lli = ll_i2info(inode);
1346 struct cl_object *obj = lli->lli_clob;
1347 struct cl_attr *attr = vvp_env_thread_attr(env);
1355 ll_inode_size_lock(inode);
1357 /* Merge timestamps the most recently obtained from MDS with
1358 * timestamps obtained from OSTs.
1360 * Do not overwrite atime of inode because it may be refreshed
1361 * by file_accessed() function. If the read was served by cache
1362 * data, there is no RPC to be sent so that atime may not be
1363 * transferred to OSTs at all. MDT only updates atime at close time
1364 * if it's at least 'mdd.*.atime_diff' older.
1365 * All in all, the atime in Lustre does not strictly comply with
1366 * POSIX. Solving this problem needs to send an RPC to MDT for each
1367 * read, this will hurt performance.
1369 if (ll_file_test_and_clear_flag(lli, LLIF_UPDATE_ATIME) ||
1370 inode->i_atime.tv_sec < lli->lli_atime)
1371 inode->i_atime.tv_sec = lli->lli_atime;
1373 inode->i_mtime.tv_sec = lli->lli_mtime;
1374 inode->i_ctime.tv_sec = lli->lli_ctime;
1376 mtime = inode->i_mtime.tv_sec;
1377 atime = inode->i_atime.tv_sec;
1378 ctime = inode->i_ctime.tv_sec;
1380 cl_object_attr_lock(obj);
1381 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1384 rc = cl_object_attr_get(env, obj, attr);
1385 cl_object_attr_unlock(obj);
1388 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1390 if (atime < attr->cat_atime)
1391 atime = attr->cat_atime;
1393 if (ctime < attr->cat_ctime)
1394 ctime = attr->cat_ctime;
1396 if (mtime < attr->cat_mtime)
1397 mtime = attr->cat_mtime;
1399 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1400 PFID(&lli->lli_fid), attr->cat_size);
1402 i_size_write(inode, attr->cat_size);
1403 inode->i_blocks = attr->cat_blocks;
1405 inode->i_mtime.tv_sec = mtime;
1406 inode->i_atime.tv_sec = atime;
1407 inode->i_ctime.tv_sec = ctime;
1410 ll_inode_size_unlock(inode);
1416 * Set designated mirror for I/O.
1418 * So far only read, write, and truncated can support to issue I/O to
1419 * designated mirror.
1421 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1423 struct ll_file_data *fd = file->private_data;
1425 /* clear layout version for generic(non-resync) I/O in case it carries
1426 * stale layout version due to I/O restart */
1427 io->ci_layout_version = 0;
1429 /* FLR: disable non-delay for designated mirror I/O because obviously
1430 * only one mirror is available */
1431 if (fd->fd_designated_mirror > 0) {
1433 io->ci_designated_mirror = fd->fd_designated_mirror;
1434 io->ci_layout_version = fd->fd_layout_version;
1437 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1438 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1441 static bool file_is_noatime(const struct file *file)
1443 const struct vfsmount *mnt = file->f_path.mnt;
1444 const struct inode *inode = file_inode((struct file *)file);
1446 /* Adapted from file_accessed() and touch_atime().*/
1447 if (file->f_flags & O_NOATIME)
1450 if (inode->i_flags & S_NOATIME)
1453 if (IS_NOATIME(inode))
1456 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1459 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1462 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1468 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1469 struct vvp_io_args *args)
1471 struct inode *inode = file_inode(file);
1472 struct ll_file_data *fd = file->private_data;
1474 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1475 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1477 if (iot == CIT_WRITE) {
1478 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1479 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1480 file->f_flags & O_DIRECT ||
1482 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1483 io->u.ci_wr.wr_sync |= !!(args &&
1484 args->via_io_subtype == IO_NORMAL &&
1485 args->u.normal.via_iocb->ki_flags & IOCB_DSYNC);
1489 io->ci_obj = ll_i2info(inode)->lli_clob;
1490 io->ci_lockreq = CILR_MAYBE;
1491 if (ll_file_nolock(file)) {
1492 io->ci_lockreq = CILR_NEVER;
1493 io->ci_no_srvlock = 1;
1494 } else if (file->f_flags & O_APPEND) {
1495 io->ci_lockreq = CILR_MANDATORY;
1497 io->ci_noatime = file_is_noatime(file);
1498 io->ci_async_readahead = false;
1500 /* FLR: only use non-delay I/O for read as there is only one
1501 * avaliable mirror for write. */
1502 io->ci_ndelay = !(iot == CIT_WRITE);
1504 ll_io_set_mirror(io, file);
1507 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1510 struct ll_inode_info *lli = ll_i2info(inode);
1511 struct ll_sb_info *sbi = ll_i2sbi(inode);
1512 enum obd_heat_type sample_type;
1513 enum obd_heat_type iobyte_type;
1514 __u64 now = ktime_get_real_seconds();
1516 if (!ll_sbi_has_file_heat(sbi) ||
1517 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1520 if (iot == CIT_READ) {
1521 sample_type = OBD_HEAT_READSAMPLE;
1522 iobyte_type = OBD_HEAT_READBYTE;
1523 } else if (iot == CIT_WRITE) {
1524 sample_type = OBD_HEAT_WRITESAMPLE;
1525 iobyte_type = OBD_HEAT_WRITEBYTE;
1530 spin_lock(&lli->lli_heat_lock);
1531 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1532 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1533 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1534 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1535 spin_unlock(&lli->lli_heat_lock);
1539 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1540 struct file *file, enum cl_io_type iot,
1541 loff_t *ppos, size_t count)
1543 struct vvp_io *vio = vvp_env_io(env);
1544 struct inode *inode = file_inode(file);
1545 struct ll_inode_info *lli = ll_i2info(inode);
1546 struct ll_file_data *fd = file->private_data;
1547 struct range_lock range;
1551 unsigned int retried = 0, ignore_lockless = 0;
1552 bool is_aio = false;
1553 struct cl_dio_aio *ci_aio = NULL;
1557 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1558 file_dentry(file)->d_name.name,
1559 iot == CIT_READ ? "read" : "write", *ppos, count);
1561 io = vvp_env_thread_io(env);
1562 if (args->via_io_subtype == IO_NORMAL &&
1563 file->f_flags & O_DIRECT) {
1564 if (!is_sync_kiocb(args->u.normal.via_iocb))
1566 ci_aio = cl_aio_alloc(args->u.normal.via_iocb);
1568 GOTO(out, rc = -ENOMEM);
1572 io = vvp_env_thread_io(env);
1573 ll_io_init(io, file, iot, args);
1574 io->ci_aio = ci_aio;
1575 io->ci_ignore_lockless = ignore_lockless;
1576 io->ci_ndelay_tried = retried;
1578 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
1579 bool range_locked = false;
1581 if (file->f_flags & O_APPEND)
1582 range_lock_init(&range, 0, LUSTRE_EOF);
1584 range_lock_init(&range, *ppos, *ppos + count - 1);
1586 vio->vui_fd = file->private_data;
1587 vio->vui_io_subtype = args->via_io_subtype;
1589 switch (vio->vui_io_subtype) {
1591 vio->vui_iter = args->u.normal.via_iter;
1592 vio->vui_iocb = args->u.normal.via_iocb;
1593 /* Direct IO reads must also take range lock,
1594 * or multiple reads will try to work on the same pages
1595 * See LU-6227 for details. */
1596 if (((iot == CIT_WRITE) ||
1597 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1598 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1599 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1601 rc = range_lock(&lli->lli_write_tree, &range);
1605 range_locked = true;
1609 vio->u.splice.vui_pipe = args->u.splice.via_pipe;
1610 vio->u.splice.vui_flags = args->u.splice.via_flags;
1613 CERROR("unknown IO subtype %u\n", vio->vui_io_subtype);
1617 ll_cl_add(file, env, io, LCC_RW);
1618 rc = cl_io_loop(env, io);
1619 ll_cl_remove(file, env);
1622 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1624 range_unlock(&lli->lli_write_tree, &range);
1627 /* cl_io_rw_init() handled IO */
1632 * In order to move forward AIO, ci_nob was increased,
1633 * but that doesn't mean io have been finished, it just
1634 * means io have been submited, we will always return
1635 * EIOCBQUEUED to the caller, So we could only return
1636 * number of bytes in non-AIO case.
1638 if (io->ci_nob > 0) {
1640 result += io->ci_nob;
1641 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1643 count -= io->ci_nob;
1645 /* prepare IO restart */
1646 if (count > 0 && args->via_io_subtype == IO_NORMAL)
1647 args->u.normal.via_iter = vio->vui_iter;
1650 cl_io_fini(env, io);
1653 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1654 file->f_path.dentry->d_name.name,
1655 iot, rc, result, io->ci_need_restart);
1657 if ((rc == 0 || rc == -ENODATA || rc == -ENOLCK) &&
1658 count > 0 && io->ci_need_restart) {
1660 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1661 file_dentry(file)->d_name.name,
1662 iot == CIT_READ ? "read" : "write",
1663 *ppos, count, result, rc);
1664 /* preserve the tried count for FLR */
1665 retried = io->ci_ndelay_tried;
1666 ignore_lockless = io->ci_ignore_lockless;
1672 * VFS will call aio_complete() if no -EIOCBQUEUED
1673 * is returned for AIO, so we can not call aio_complete()
1676 if (rc != -EIOCBQUEUED)
1677 io->ci_aio->cda_no_aio_complete = 1;
1679 * Drop one extra reference so that end_io() could be
1680 * called for this IO context, we could call it after
1681 * we make sure all AIO requests have been proceed.
1683 cl_sync_io_note(env, &io->ci_aio->cda_sync,
1684 rc == -EIOCBQUEUED ? 0 : rc);
1686 cl_aio_free(io->ci_aio);
1691 if (iot == CIT_READ) {
1693 ll_stats_ops_tally(ll_i2sbi(inode),
1694 LPROC_LL_READ_BYTES, result);
1695 } else if (iot == CIT_WRITE) {
1697 ll_stats_ops_tally(ll_i2sbi(inode),
1698 LPROC_LL_WRITE_BYTES, result);
1699 fd->fd_write_failed = false;
1700 } else if (result == 0 && rc == 0) {
1703 fd->fd_write_failed = true;
1705 fd->fd_write_failed = false;
1706 } else if (rc != -ERESTARTSYS) {
1707 fd->fd_write_failed = true;
1711 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1713 ll_heat_add(inode, iot, result);
1715 RETURN(result > 0 ? result : rc);
1719 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1720 * especially for small I/O.
1722 * To serve a read request, CLIO has to create and initialize a cl_io and
1723 * then request DLM lock. This has turned out to have siginificant overhead
1724 * and affects the performance of small I/O dramatically.
1726 * It's not necessary to create a cl_io for each I/O. Under the help of read
1727 * ahead, most of the pages being read are already in memory cache and we can
1728 * read those pages directly because if the pages exist, the corresponding DLM
1729 * lock must exist so that page content must be valid.
1731 * In fast read implementation, the llite speculatively finds and reads pages
1732 * in memory cache. There are three scenarios for fast read:
1733 * - If the page exists and is uptodate, kernel VM will provide the data and
1734 * CLIO won't be intervened;
1735 * - If the page was brought into memory by read ahead, it will be exported
1736 * and read ahead parameters will be updated;
1737 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1738 * it will go back and invoke normal read, i.e., a cl_io will be created
1739 * and DLM lock will be requested.
1741 * POSIX compliance: posix standard states that read is intended to be atomic.
1742 * Lustre read implementation is in line with Linux kernel read implementation
1743 * and neither of them complies with POSIX standard in this matter. Fast read
1744 * doesn't make the situation worse on single node but it may interleave write
1745 * results from multiple nodes due to short read handling in ll_file_aio_read().
1747 * \param env - lu_env
1748 * \param iocb - kiocb from kernel
1749 * \param iter - user space buffers where the data will be copied
1751 * \retval - number of bytes have been read, or error code if error occurred.
1754 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1758 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1761 /* NB: we can't do direct IO for fast read because it will need a lock
1762 * to make IO engine happy. */
1763 if (iocb->ki_filp->f_flags & O_DIRECT)
1766 result = generic_file_read_iter(iocb, iter);
1768 /* If the first page is not in cache, generic_file_aio_read() will be
1769 * returned with -ENODATA.
1770 * See corresponding code in ll_readpage(). */
1771 if (result == -ENODATA)
1775 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1776 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1777 LPROC_LL_READ_BYTES, result);
1784 * Read from a file (through the page cache).
1786 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1789 struct vvp_io_args *args;
1790 struct file *file = iocb->ki_filp;
1794 ktime_t kstart = ktime_get();
1797 if (!iov_iter_count(to))
1801 * Currently when PCC read failed, we do not fall back to the
1802 * normal read path, just return the error.
1803 * The resaon is that: for RW-PCC, the file data may be modified
1804 * in the PCC and inconsistent with the data on OSTs (or file
1805 * data has been removed from the Lustre file system), at this
1806 * time, fallback to the normal read path may read the wrong
1808 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1809 * path: read data from data copy on OSTs.
1811 result = pcc_file_read_iter(iocb, to, &cached);
1815 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
1817 result = ll_do_fast_read(iocb, to);
1818 if (result < 0 || iov_iter_count(to) == 0)
1821 env = cl_env_get(&refcheck);
1823 return PTR_ERR(env);
1825 args = ll_env_args(env, IO_NORMAL);
1826 args->u.normal.via_iter = to;
1827 args->u.normal.via_iocb = iocb;
1829 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
1830 &iocb->ki_pos, iov_iter_count(to));
1833 else if (result == 0)
1836 cl_env_put(env, &refcheck);
1839 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1840 file->private_data, iocb->ki_pos, result,
1842 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
1843 ktime_us_delta(ktime_get(), kstart));
1850 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
1851 * If a page is already in the page cache and dirty (and some other things -
1852 * See ll_tiny_write_begin for the instantiation of these rules), then we can
1853 * write to it without doing a full I/O, because Lustre already knows about it
1854 * and will write it out. This saves a lot of processing time.
1856 * All writes here are within one page, so exclusion is handled by the page
1857 * lock on the vm page. We do not do tiny writes for writes which touch
1858 * multiple pages because it's very unlikely multiple sequential pages are
1859 * are already dirty.
1861 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
1862 * and are unlikely to be to already dirty pages.
1864 * Attribute updates are important here, we do them in ll_tiny_write_end.
1866 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
1868 ssize_t count = iov_iter_count(iter);
1869 struct file *file = iocb->ki_filp;
1870 struct inode *inode = file_inode(file);
1871 bool lock_inode = !IS_NOSEC(inode);
1876 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
1877 * of function for why.
1879 if (count >= PAGE_SIZE ||
1880 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
1883 if (unlikely(lock_inode))
1885 result = __generic_file_write_iter(iocb, iter);
1887 if (unlikely(lock_inode))
1888 inode_unlock(inode);
1890 /* If the page is not already dirty, ll_tiny_write_begin returns
1891 * -ENODATA. We continue on to normal write.
1893 if (result == -ENODATA)
1897 ll_heat_add(inode, CIT_WRITE, result);
1898 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES,
1900 ll_file_set_flag(ll_i2info(inode), LLIF_DATA_MODIFIED);
1903 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
1909 * Write to a file (through the page cache).
1911 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1913 struct vvp_io_args *args;
1915 ssize_t rc_tiny = 0, rc_normal;
1916 struct file *file = iocb->ki_filp;
1919 ktime_t kstart = ktime_get();
1924 if (!iov_iter_count(from))
1925 GOTO(out, rc_normal = 0);
1928 * When PCC write failed, we usually do not fall back to the normal
1929 * write path, just return the error. But there is a special case when
1930 * returned error code is -ENOSPC due to running out of space on PCC HSM
1931 * bakcend. At this time, it will fall back to normal I/O path and
1932 * retry the I/O. As the file is in HSM released state, it will restore
1933 * the file data to OSTs first and redo the write again. And the
1934 * restore process will revoke the layout lock and detach the file
1935 * from PCC cache automatically.
1937 result = pcc_file_write_iter(iocb, from, &cached);
1938 if (cached && result != -ENOSPC && result != -EDQUOT)
1939 GOTO(out, rc_normal = result);
1941 /* NB: we can't do direct IO for tiny writes because they use the page
1942 * cache, we can't do sync writes because tiny writes can't flush
1943 * pages, and we can't do append writes because we can't guarantee the
1944 * required DLM locks are held to protect file size.
1946 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
1947 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
1948 rc_tiny = ll_do_tiny_write(iocb, from);
1950 /* In case of error, go on and try normal write - Only stop if tiny
1951 * write completed I/O.
1953 if (iov_iter_count(from) == 0)
1954 GOTO(out, rc_normal = rc_tiny);
1956 env = cl_env_get(&refcheck);
1958 return PTR_ERR(env);
1960 args = ll_env_args(env, IO_NORMAL);
1961 args->u.normal.via_iter = from;
1962 args->u.normal.via_iocb = iocb;
1964 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
1965 &iocb->ki_pos, iov_iter_count(from));
1967 /* On success, combine bytes written. */
1968 if (rc_tiny >= 0 && rc_normal > 0)
1969 rc_normal += rc_tiny;
1970 /* On error, only return error from normal write if tiny write did not
1971 * write any bytes. Otherwise return bytes written by tiny write.
1973 else if (rc_tiny > 0)
1974 rc_normal = rc_tiny;
1976 cl_env_put(env, &refcheck);
1978 if (rc_normal > 0) {
1979 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1980 file->private_data, iocb->ki_pos,
1982 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
1983 ktime_us_delta(ktime_get(), kstart));
1989 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1991 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
1993 static int ll_file_get_iov_count(const struct iovec *iov,
1994 unsigned long *nr_segs, size_t *count,
2000 for (seg = 0; seg < *nr_segs; seg++) {
2001 const struct iovec *iv = &iov[seg];
2004 * If any segment has a negative length, or the cumulative
2005 * length ever wraps negative then return -EINVAL.
2008 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2010 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2015 cnt -= iv->iov_len; /* This segment is no good */
2022 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2023 unsigned long nr_segs, loff_t pos)
2030 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2037 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2038 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2039 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2040 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2041 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2043 result = ll_file_read_iter(iocb, &to);
2048 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2051 struct iovec iov = { .iov_base = buf, .iov_len = count };
2060 init_sync_kiocb(&kiocb, file);
2061 kiocb.ki_pos = *ppos;
2062 #ifdef HAVE_KIOCB_KI_LEFT
2063 kiocb.ki_left = count;
2064 #elif defined(HAVE_KI_NBYTES)
2065 kiocb.i_nbytes = count;
2068 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2069 *ppos = kiocb.ki_pos;
2075 * Write to a file (through the page cache).
2078 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2079 unsigned long nr_segs, loff_t pos)
2081 struct iov_iter from;
2086 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2093 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2094 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2095 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2096 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2097 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2099 result = ll_file_write_iter(iocb, &from);
2104 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2105 size_t count, loff_t *ppos)
2107 struct iovec iov = { .iov_base = (void __user *)buf,
2117 init_sync_kiocb(&kiocb, file);
2118 kiocb.ki_pos = *ppos;
2119 #ifdef HAVE_KIOCB_KI_LEFT
2120 kiocb.ki_left = count;
2121 #elif defined(HAVE_KI_NBYTES)
2122 kiocb.ki_nbytes = count;
2125 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2126 *ppos = kiocb.ki_pos;
2130 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2133 * Send file content (through pagecache) somewhere with helper
2135 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
2136 struct pipe_inode_info *pipe, size_t count,
2140 struct vvp_io_args *args;
2147 result = pcc_file_splice_read(in_file, ppos, pipe,
2148 count, flags, &cached);
2152 ll_ras_enter(in_file, *ppos, count);
2154 env = cl_env_get(&refcheck);
2156 RETURN(PTR_ERR(env));
2158 args = ll_env_args(env, IO_SPLICE);
2159 args->u.splice.via_pipe = pipe;
2160 args->u.splice.via_flags = flags;
2162 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
2163 cl_env_put(env, &refcheck);
2166 ll_rw_stats_tally(ll_i2sbi(file_inode(in_file)), current->pid,
2167 in_file->private_data, *ppos, result,
2172 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2173 __u64 flags, struct lov_user_md *lum, int lum_size)
2175 struct lookup_intent oit = {
2177 .it_flags = flags | MDS_OPEN_BY_FID,
2182 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2183 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2184 /* this code will only exist for big-endian systems */
2185 lustre_swab_lov_user_md(lum, 0);
2188 ll_inode_size_lock(inode);
2189 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2191 GOTO(out_unlock, rc);
2193 ll_release_openhandle(dentry, &oit);
2196 ll_inode_size_unlock(inode);
2197 ll_intent_release(&oit);
2202 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2203 struct lov_mds_md **lmmp, int *lmm_size,
2204 struct ptlrpc_request **request)
2206 struct ll_sb_info *sbi = ll_i2sbi(inode);
2207 struct mdt_body *body;
2208 struct lov_mds_md *lmm = NULL;
2209 struct ptlrpc_request *req = NULL;
2210 struct md_op_data *op_data;
2213 rc = ll_get_default_mdsize(sbi, &lmmsize);
2217 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2218 strlen(filename), lmmsize,
2219 LUSTRE_OPC_ANY, NULL);
2220 if (IS_ERR(op_data))
2221 RETURN(PTR_ERR(op_data));
2223 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2224 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2225 ll_finish_md_op_data(op_data);
2227 CDEBUG(D_INFO, "md_getattr_name failed "
2228 "on %s: rc %d\n", filename, rc);
2232 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2233 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2235 lmmsize = body->mbo_eadatasize;
2237 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2239 GOTO(out, rc = -ENODATA);
2242 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2243 LASSERT(lmm != NULL);
2245 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2246 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2247 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2248 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2249 GOTO(out, rc = -EPROTO);
2252 * This is coming from the MDS, so is probably in
2253 * little endian. We convert it to host endian before
2254 * passing it to userspace.
2256 if ((lmm->lmm_magic & __swab32(LOV_MAGIC_MAGIC)) ==
2257 __swab32(LOV_MAGIC_MAGIC)) {
2258 int stripe_count = 0;
2260 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2261 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2262 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2263 if (le32_to_cpu(lmm->lmm_pattern) &
2264 LOV_PATTERN_F_RELEASED)
2268 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2270 /* if function called for directory - we should
2271 * avoid swab not existent lsm objects */
2272 if (lmm->lmm_magic == LOV_MAGIC_V1 && S_ISREG(body->mbo_mode))
2273 lustre_swab_lov_user_md_objects(
2274 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2276 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2277 S_ISREG(body->mbo_mode))
2278 lustre_swab_lov_user_md_objects(
2279 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2285 *lmm_size = lmmsize;
2290 static int ll_lov_setea(struct inode *inode, struct file *file,
2293 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2294 struct lov_user_md *lump;
2295 int lum_size = sizeof(struct lov_user_md) +
2296 sizeof(struct lov_user_ost_data);
2300 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2303 OBD_ALLOC_LARGE(lump, lum_size);
2307 if (copy_from_user(lump, arg, lum_size))
2308 GOTO(out_lump, rc = -EFAULT);
2310 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2312 cl_lov_delay_create_clear(&file->f_flags);
2315 OBD_FREE_LARGE(lump, lum_size);
2319 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2326 env = cl_env_get(&refcheck);
2328 RETURN(PTR_ERR(env));
2330 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2331 cl_env_put(env, &refcheck);
2335 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2338 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2339 struct lov_user_md *klum;
2341 __u64 flags = FMODE_WRITE;
2344 rc = ll_copy_user_md(lum, &klum);
2349 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2354 rc = put_user(0, &lum->lmm_stripe_count);
2358 rc = ll_layout_refresh(inode, &gen);
2362 rc = ll_file_getstripe(inode, arg, lum_size);
2363 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2364 ll_i2info(inode)->lli_clob) {
2365 struct iattr attr = { 0 };
2367 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2368 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2371 cl_lov_delay_create_clear(&file->f_flags);
2374 OBD_FREE_LARGE(klum, lum_size);
2380 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2382 struct ll_inode_info *lli = ll_i2info(inode);
2383 struct cl_object *obj = lli->lli_clob;
2384 struct ll_file_data *fd = file->private_data;
2385 struct ll_grouplock grouplock;
2390 CWARN("group id for group lock must not be 0\n");
2394 if (ll_file_nolock(file))
2395 RETURN(-EOPNOTSUPP);
2397 if (file->f_flags & O_NONBLOCK) {
2398 if (!mutex_trylock(&lli->lli_group_mutex))
2401 mutex_lock(&lli->lli_group_mutex);
2403 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2404 CWARN("group lock already existed with gid %lu\n",
2405 fd->fd_grouplock.lg_gid);
2406 GOTO(out, rc = -EINVAL);
2408 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2409 if (file->f_flags & O_NONBLOCK)
2410 GOTO(out, rc = -EAGAIN);
2411 mutex_unlock(&lli->lli_group_mutex);
2412 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2413 GOTO(retry, rc = 0);
2415 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2418 * XXX: group lock needs to protect all OST objects while PFL
2419 * can add new OST objects during the IO, so we'd instantiate
2420 * all OST objects before getting its group lock.
2425 struct cl_layout cl = {
2426 .cl_is_composite = false,
2428 struct lu_extent ext = {
2430 .e_end = OBD_OBJECT_EOF,
2433 env = cl_env_get(&refcheck);
2435 GOTO(out, rc = PTR_ERR(env));
2437 rc = cl_object_layout_get(env, obj, &cl);
2438 if (!rc && cl.cl_is_composite)
2439 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2442 cl_env_put(env, &refcheck);
2447 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2448 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2453 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2454 fd->fd_grouplock = grouplock;
2455 if (lli->lli_group_users == 0)
2456 lli->lli_group_gid = grouplock.lg_gid;
2457 lli->lli_group_users++;
2459 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2461 mutex_unlock(&lli->lli_group_mutex);
2466 static int ll_put_grouplock(struct inode *inode, struct file *file,
2469 struct ll_inode_info *lli = ll_i2info(inode);
2470 struct ll_file_data *fd = file->private_data;
2471 struct ll_grouplock grouplock;
2475 mutex_lock(&lli->lli_group_mutex);
2476 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2477 CWARN("no group lock held\n");
2478 GOTO(out, rc = -EINVAL);
2481 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2483 if (fd->fd_grouplock.lg_gid != arg) {
2484 CWARN("group lock %lu doesn't match current id %lu\n",
2485 arg, fd->fd_grouplock.lg_gid);
2486 GOTO(out, rc = -EINVAL);
2489 grouplock = fd->fd_grouplock;
2490 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2491 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2493 cl_put_grouplock(&grouplock);
2495 lli->lli_group_users--;
2496 if (lli->lli_group_users == 0) {
2497 lli->lli_group_gid = 0;
2498 wake_up_var(&lli->lli_group_users);
2500 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2503 mutex_unlock(&lli->lli_group_mutex);
2509 * Close inode open handle
2511 * \param dentry [in] dentry which contains the inode
2512 * \param it [in,out] intent which contains open info and result
2515 * \retval <0 failure
2517 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2519 struct inode *inode = dentry->d_inode;
2520 struct obd_client_handle *och;
2526 /* Root ? Do nothing. */
2527 if (dentry->d_inode->i_sb->s_root == dentry)
2530 /* No open handle to close? Move away */
2531 if (!it_disposition(it, DISP_OPEN_OPEN))
2534 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2536 OBD_ALLOC(och, sizeof(*och));
2538 GOTO(out, rc = -ENOMEM);
2540 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2544 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2546 /* this one is in place of ll_file_open */
2547 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2548 ptlrpc_req_finished(it->it_request);
2549 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2555 * Get size for inode for which FIEMAP mapping is requested.
2556 * Make the FIEMAP get_info call and returns the result.
2557 * \param fiemap kernel buffer to hold extens
2558 * \param num_bytes kernel buffer size
2560 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2566 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2569 /* Checks for fiemap flags */
2570 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2571 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2575 /* Check for FIEMAP_FLAG_SYNC */
2576 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2577 rc = filemap_fdatawrite(inode->i_mapping);
2582 env = cl_env_get(&refcheck);
2584 RETURN(PTR_ERR(env));
2586 if (i_size_read(inode) == 0) {
2587 rc = ll_glimpse_size(inode);
2592 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2593 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2594 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2596 /* If filesize is 0, then there would be no objects for mapping */
2597 if (fmkey.lfik_oa.o_size == 0) {
2598 fiemap->fm_mapped_extents = 0;
2602 fmkey.lfik_fiemap = *fiemap;
2604 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2605 &fmkey, fiemap, &num_bytes);
2607 cl_env_put(env, &refcheck);
2611 int ll_fid2path(struct inode *inode, void __user *arg)
2613 struct obd_export *exp = ll_i2mdexp(inode);
2614 const struct getinfo_fid2path __user *gfin = arg;
2616 struct getinfo_fid2path *gfout;
2622 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2623 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2626 /* Only need to get the buflen */
2627 if (get_user(pathlen, &gfin->gf_pathlen))
2630 if (pathlen > PATH_MAX)
2633 outsize = sizeof(*gfout) + pathlen;
2634 OBD_ALLOC(gfout, outsize);
2638 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2639 GOTO(gf_free, rc = -EFAULT);
2640 /* append root FID after gfout to let MDT know the root FID so that it
2641 * can lookup the correct path, this is mainly for fileset.
2642 * old server without fileset mount support will ignore this. */
2643 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2645 /* Call mdc_iocontrol */
2646 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2650 if (copy_to_user(arg, gfout, outsize))
2654 OBD_FREE(gfout, outsize);
2659 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2661 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2669 ioc->idv_version = 0;
2670 ioc->idv_layout_version = UINT_MAX;
2672 /* If no file object initialized, we consider its version is 0. */
2676 env = cl_env_get(&refcheck);
2678 RETURN(PTR_ERR(env));
2680 io = vvp_env_thread_io(env);
2682 io->u.ci_data_version.dv_data_version = 0;
2683 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2684 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2687 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2688 result = cl_io_loop(env, io);
2690 result = io->ci_result;
2692 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2693 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2695 cl_io_fini(env, io);
2697 if (unlikely(io->ci_need_restart))
2700 cl_env_put(env, &refcheck);
2706 * Read the data_version for inode.
2708 * This value is computed using stripe object version on OST.
2709 * Version is computed using server side locking.
2711 * @param flags if do sync on the OST side;
2713 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2714 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2716 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2718 struct ioc_data_version ioc = { .idv_flags = flags };
2721 rc = ll_ioc_data_version(inode, &ioc);
2723 *data_version = ioc.idv_version;
2729 * Trigger a HSM release request for the provided inode.
2731 int ll_hsm_release(struct inode *inode)
2734 struct obd_client_handle *och = NULL;
2735 __u64 data_version = 0;
2740 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2741 ll_i2sbi(inode)->ll_fsname,
2742 PFID(&ll_i2info(inode)->lli_fid));
2744 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2746 GOTO(out, rc = PTR_ERR(och));
2748 /* Grab latest data_version and [am]time values */
2749 rc = ll_data_version(inode, &data_version, LL_DV_WR_FLUSH);
2753 env = cl_env_get(&refcheck);
2755 GOTO(out, rc = PTR_ERR(env));
2757 rc = ll_merge_attr(env, inode);
2758 cl_env_put(env, &refcheck);
2760 /* If error happen, we have the wrong size for a file.
2766 /* Release the file.
2767 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2768 * we still need it to pack l_remote_handle to MDT. */
2769 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2775 if (och != NULL && !IS_ERR(och)) /* close the file */
2776 ll_lease_close(och, inode, NULL);
2781 struct ll_swap_stack {
2784 struct inode *inode1;
2785 struct inode *inode2;
2790 static int ll_swap_layouts(struct file *file1, struct file *file2,
2791 struct lustre_swap_layouts *lsl)
2793 struct mdc_swap_layouts msl;
2794 struct md_op_data *op_data;
2797 struct ll_swap_stack *llss = NULL;
2800 OBD_ALLOC_PTR(llss);
2804 llss->inode1 = file_inode(file1);
2805 llss->inode2 = file_inode(file2);
2807 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2811 /* we use 2 bool because it is easier to swap than 2 bits */
2812 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2813 llss->check_dv1 = true;
2815 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2816 llss->check_dv2 = true;
2818 /* we cannot use lsl->sl_dvX directly because we may swap them */
2819 llss->dv1 = lsl->sl_dv1;
2820 llss->dv2 = lsl->sl_dv2;
2822 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2823 if (rc == 0) /* same file, done! */
2826 if (rc < 0) { /* sequentialize it */
2827 swap(llss->inode1, llss->inode2);
2829 swap(llss->dv1, llss->dv2);
2830 swap(llss->check_dv1, llss->check_dv2);
2834 if (gid != 0) { /* application asks to flush dirty cache */
2835 rc = ll_get_grouplock(llss->inode1, file1, gid);
2839 rc = ll_get_grouplock(llss->inode2, file2, gid);
2841 ll_put_grouplock(llss->inode1, file1, gid);
2846 /* ultimate check, before swaping the layouts we check if
2847 * dataversion has changed (if requested) */
2848 if (llss->check_dv1) {
2849 rc = ll_data_version(llss->inode1, &dv, 0);
2852 if (dv != llss->dv1)
2853 GOTO(putgl, rc = -EAGAIN);
2856 if (llss->check_dv2) {
2857 rc = ll_data_version(llss->inode2, &dv, 0);
2860 if (dv != llss->dv2)
2861 GOTO(putgl, rc = -EAGAIN);
2864 /* struct md_op_data is used to send the swap args to the mdt
2865 * only flags is missing, so we use struct mdc_swap_layouts
2866 * through the md_op_data->op_data */
2867 /* flags from user space have to be converted before they are send to
2868 * server, no flag is sent today, they are only used on the client */
2871 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
2872 0, LUSTRE_OPC_ANY, &msl);
2873 if (IS_ERR(op_data))
2874 GOTO(free, rc = PTR_ERR(op_data));
2876 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
2877 sizeof(*op_data), op_data, NULL);
2878 ll_finish_md_op_data(op_data);
2885 ll_put_grouplock(llss->inode2, file2, gid);
2886 ll_put_grouplock(llss->inode1, file1, gid);
2896 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
2898 struct obd_export *exp = ll_i2mdexp(inode);
2899 struct md_op_data *op_data;
2903 /* Detect out-of range masks */
2904 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
2907 /* Non-root users are forbidden to set or clear flags which are
2908 * NOT defined in HSM_USER_MASK. */
2909 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
2910 !cfs_capable(CFS_CAP_SYS_ADMIN))
2913 if (!exp_connect_archive_id_array(exp)) {
2914 /* Detect out-of range archive id */
2915 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
2916 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
2920 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2921 LUSTRE_OPC_ANY, hss);
2922 if (IS_ERR(op_data))
2923 RETURN(PTR_ERR(op_data));
2925 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
2928 ll_finish_md_op_data(op_data);
2933 static int ll_hsm_import(struct inode *inode, struct file *file,
2934 struct hsm_user_import *hui)
2936 struct hsm_state_set *hss = NULL;
2937 struct iattr *attr = NULL;
2941 if (!S_ISREG(inode->i_mode))
2947 GOTO(out, rc = -ENOMEM);
2949 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
2950 hss->hss_archive_id = hui->hui_archive_id;
2951 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
2952 rc = ll_hsm_state_set(inode, hss);
2956 OBD_ALLOC_PTR(attr);
2958 GOTO(out, rc = -ENOMEM);
2960 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
2961 attr->ia_mode |= S_IFREG;
2962 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
2963 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
2964 attr->ia_size = hui->hui_size;
2965 attr->ia_mtime.tv_sec = hui->hui_mtime;
2966 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
2967 attr->ia_atime.tv_sec = hui->hui_atime;
2968 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
2970 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
2971 ATTR_UID | ATTR_GID |
2972 ATTR_MTIME | ATTR_MTIME_SET |
2973 ATTR_ATIME | ATTR_ATIME_SET;
2977 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
2981 inode_unlock(inode);
2993 static inline long ll_lease_type_from_fmode(fmode_t fmode)
2995 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
2996 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
2999 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3001 struct inode *inode = file_inode(file);
3003 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3004 ATTR_MTIME | ATTR_MTIME_SET |
3007 .tv_sec = lfu->lfu_atime_sec,
3008 .tv_nsec = lfu->lfu_atime_nsec,
3011 .tv_sec = lfu->lfu_mtime_sec,
3012 .tv_nsec = lfu->lfu_mtime_nsec,
3015 .tv_sec = lfu->lfu_ctime_sec,
3016 .tv_nsec = lfu->lfu_ctime_nsec,
3022 if (!capable(CAP_SYS_ADMIN))
3025 if (!S_ISREG(inode->i_mode))
3029 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3031 inode_unlock(inode);
3036 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3039 case MODE_READ_USER:
3041 case MODE_WRITE_USER:
3048 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3050 /* Used to allow the upper layers of the client to request an LDLM lock
3051 * without doing an actual read or write.
3053 * Used for ladvise lockahead to manually request specific locks.
3055 * \param[in] file file this ladvise lock request is on
3056 * \param[in] ladvise ladvise struct describing this lock request
3058 * \retval 0 success, no detailed result available (sync requests
3059 * and requests sent to the server [not handled locally]
3060 * cannot return detailed results)
3061 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3062 * see definitions for details.
3063 * \retval negative negative errno on error
3065 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3067 struct lu_env *env = NULL;
3068 struct cl_io *io = NULL;
3069 struct cl_lock *lock = NULL;
3070 struct cl_lock_descr *descr = NULL;
3071 struct dentry *dentry = file->f_path.dentry;
3072 struct inode *inode = dentry->d_inode;
3073 enum cl_lock_mode cl_mode;
3074 off_t start = ladvise->lla_start;
3075 off_t end = ladvise->lla_end;
3082 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3083 dentry, dentry->d_inode,
3084 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3087 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3089 GOTO(out, result = cl_mode);
3091 /* Get IO environment */
3092 result = cl_io_get(inode, &env, &io, &refcheck);
3096 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3099 * nothing to do for this io. This currently happens when
3100 * stripe sub-object's are not yet created.
3102 result = io->ci_result;
3103 } else if (result == 0) {
3104 lock = vvp_env_lock(env);
3105 descr = &lock->cll_descr;
3107 descr->cld_obj = io->ci_obj;
3108 /* Convert byte offsets to pages */
3109 descr->cld_start = cl_index(io->ci_obj, start);
3110 descr->cld_end = cl_index(io->ci_obj, end);
3111 descr->cld_mode = cl_mode;
3112 /* CEF_MUST is used because we do not want to convert a
3113 * lockahead request to a lockless lock */
3114 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND |
3117 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3118 descr->cld_enq_flags |= CEF_SPECULATIVE;
3120 result = cl_lock_request(env, io, lock);
3122 /* On success, we need to release the lock */
3124 cl_lock_release(env, lock);
3126 cl_io_fini(env, io);
3127 cl_env_put(env, &refcheck);
3129 /* -ECANCELED indicates a matching lock with a different extent
3130 * was already present, and -EEXIST indicates a matching lock
3131 * on exactly the same extent was already present.
3132 * We convert them to positive values for userspace to make
3133 * recognizing true errors easier.
3134 * Note we can only return these detailed results on async requests,
3135 * as sync requests look the same as i/o requests for locking. */
3136 if (result == -ECANCELED)
3137 result = LLA_RESULT_DIFFERENT;
3138 else if (result == -EEXIST)
3139 result = LLA_RESULT_SAME;
3144 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3146 static int ll_ladvise_sanity(struct inode *inode,
3147 struct llapi_lu_ladvise *ladvise)
3149 struct ll_sb_info *sbi = ll_i2sbi(inode);
3150 enum lu_ladvise_type advice = ladvise->lla_advice;
3151 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3152 * be in the first 32 bits of enum ladvise_flags */
3153 __u32 flags = ladvise->lla_peradvice_flags;
3154 /* 3 lines at 80 characters per line, should be plenty */
3157 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3160 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3161 sbi->ll_fsname, advice,
3162 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3166 /* Per-advice checks */
3168 case LU_LADVISE_LOCKNOEXPAND:
3169 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3171 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3172 "rc = %d\n", sbi->ll_fsname, flags,
3173 ladvise_names[advice], rc);
3177 case LU_LADVISE_LOCKAHEAD:
3178 /* Currently only READ and WRITE modes can be requested */
3179 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3180 ladvise->lla_lockahead_mode == 0) {
3182 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3183 "rc = %d\n", sbi->ll_fsname,
3184 ladvise->lla_lockahead_mode,
3185 ladvise_names[advice], rc);
3189 case LU_LADVISE_WILLREAD:
3190 case LU_LADVISE_DONTNEED:
3192 /* Note fall through above - These checks apply to all advices
3193 * except LOCKNOEXPAND */
3194 if (flags & ~LF_DEFAULT_MASK) {
3196 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3197 "rc = %d\n", sbi->ll_fsname, flags,
3198 ladvise_names[advice], rc);
3201 if (ladvise->lla_start >= ladvise->lla_end) {
3203 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3204 "for %s: rc = %d\n", sbi->ll_fsname,
3205 ladvise->lla_start, ladvise->lla_end,
3206 ladvise_names[advice], rc);
3218 * Give file access advices
3220 * The ladvise interface is similar to Linux fadvise() system call, except it
3221 * forwards the advices directly from Lustre client to server. The server side
3222 * codes will apply appropriate read-ahead and caching techniques for the
3223 * corresponding files.
3225 * A typical workload for ladvise is e.g. a bunch of different clients are
3226 * doing small random reads of a file, so prefetching pages into OSS cache
3227 * with big linear reads before the random IO is a net benefit. Fetching
3228 * all that data into each client cache with fadvise() may not be, due to
3229 * much more data being sent to the client.
3231 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3232 struct llapi_lu_ladvise *ladvise)
3236 struct cl_ladvise_io *lio;
3241 env = cl_env_get(&refcheck);
3243 RETURN(PTR_ERR(env));
3245 io = vvp_env_thread_io(env);
3246 io->ci_obj = ll_i2info(inode)->lli_clob;
3248 /* initialize parameters for ladvise */
3249 lio = &io->u.ci_ladvise;
3250 lio->li_start = ladvise->lla_start;
3251 lio->li_end = ladvise->lla_end;
3252 lio->li_fid = ll_inode2fid(inode);
3253 lio->li_advice = ladvise->lla_advice;
3254 lio->li_flags = flags;
3256 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3257 rc = cl_io_loop(env, io);
3261 cl_io_fini(env, io);
3262 cl_env_put(env, &refcheck);
3266 static int ll_lock_noexpand(struct file *file, int flags)
3268 struct ll_file_data *fd = file->private_data;
3270 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3275 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3278 struct fsxattr fsxattr;
3280 if (copy_from_user(&fsxattr,
3281 (const struct fsxattr __user *)arg,
3285 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3286 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT))
3287 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3288 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3289 if (copy_to_user((struct fsxattr __user *)arg,
3290 &fsxattr, sizeof(fsxattr)))
3296 int ll_ioctl_check_project(struct inode *inode, struct fsxattr *fa)
3299 * Project Quota ID state is only allowed to change from within the init
3300 * namespace. Enforce that restriction only if we are trying to change
3301 * the quota ID state. Everything else is allowed in user namespaces.
3303 if (current_user_ns() == &init_user_ns)
3306 if (ll_i2info(inode)->lli_projid != fa->fsx_projid)
3309 if (ll_file_test_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT)) {
3310 if (!(fa->fsx_xflags & FS_XFLAG_PROJINHERIT))
3313 if (fa->fsx_xflags & FS_XFLAG_PROJINHERIT)
3320 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3324 struct md_op_data *op_data;
3325 struct ptlrpc_request *req = NULL;
3327 struct fsxattr fsxattr;
3328 struct cl_object *obj;
3332 if (copy_from_user(&fsxattr,
3333 (const struct fsxattr __user *)arg,
3337 rc = ll_ioctl_check_project(inode, &fsxattr);
3341 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3342 LUSTRE_OPC_ANY, NULL);
3343 if (IS_ERR(op_data))
3344 RETURN(PTR_ERR(op_data));
3346 flags = ll_xflags_to_inode_flags(fsxattr.fsx_xflags);
3347 op_data->op_attr_flags = ll_inode_to_ext_flags(flags);
3348 if (fsxattr.fsx_xflags & FS_XFLAG_PROJINHERIT)
3349 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3350 op_data->op_projid = fsxattr.fsx_projid;
3351 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3352 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL,
3354 ptlrpc_req_finished(req);
3356 GOTO(out_fsxattr, rc);
3357 ll_update_inode_flags(inode, op_data->op_attr_flags);
3358 obj = ll_i2info(inode)->lli_clob;
3360 GOTO(out_fsxattr, rc);
3362 /* Avoiding OST RPC if this is only project ioctl */
3363 if (fsxattr.fsx_xflags == 0 ||
3364 fsxattr.fsx_xflags == FS_XFLAG_PROJINHERIT)
3365 GOTO(out_fsxattr, rc);
3367 OBD_ALLOC_PTR(attr);
3369 GOTO(out_fsxattr, rc = -ENOMEM);
3371 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS,
3372 fsxattr.fsx_xflags);
3375 ll_finish_md_op_data(op_data);
3379 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3382 struct inode *inode = file_inode(file);
3383 struct ll_file_data *fd = file->private_data;
3384 struct ll_inode_info *lli = ll_i2info(inode);
3385 struct obd_client_handle *och = NULL;
3386 struct split_param sp;
3387 struct pcc_param param;
3388 bool lease_broken = false;
3390 enum mds_op_bias bias = 0;
3391 struct file *layout_file = NULL;
3393 size_t data_size = 0;
3394 bool attached = false;
3399 mutex_lock(&lli->lli_och_mutex);
3400 if (fd->fd_lease_och != NULL) {
3401 och = fd->fd_lease_och;
3402 fd->fd_lease_och = NULL;
3404 mutex_unlock(&lli->lli_och_mutex);
3409 fmode = och->och_flags;
3411 switch (ioc->lil_flags) {
3412 case LL_LEASE_RESYNC_DONE:
3413 if (ioc->lil_count > IOC_IDS_MAX)
3414 GOTO(out_lease_close, rc = -EINVAL);
3416 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3417 OBD_ALLOC(data, data_size);
3419 GOTO(out_lease_close, rc = -ENOMEM);
3421 if (copy_from_user(data, (void __user *)arg, data_size))
3422 GOTO(out_lease_close, rc = -EFAULT);
3424 bias = MDS_CLOSE_RESYNC_DONE;
3426 case LL_LEASE_LAYOUT_MERGE: {
3429 if (ioc->lil_count != 1)
3430 GOTO(out_lease_close, rc = -EINVAL);
3432 arg += sizeof(*ioc);
3433 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3434 GOTO(out_lease_close, rc = -EFAULT);
3436 layout_file = fget(fd);
3438 GOTO(out_lease_close, rc = -EBADF);
3440 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3441 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3442 GOTO(out_lease_close, rc = -EPERM);
3444 data = file_inode(layout_file);
3445 bias = MDS_CLOSE_LAYOUT_MERGE;
3448 case LL_LEASE_LAYOUT_SPLIT: {
3452 if (ioc->lil_count != 2)
3453 GOTO(out_lease_close, rc = -EINVAL);
3455 arg += sizeof(*ioc);
3456 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3457 GOTO(out_lease_close, rc = -EFAULT);
3459 arg += sizeof(__u32);
3460 if (copy_from_user(&mirror_id, (void __user *)arg,
3462 GOTO(out_lease_close, rc = -EFAULT);
3464 layout_file = fget(fdv);
3466 GOTO(out_lease_close, rc = -EBADF);
3468 sp.sp_inode = file_inode(layout_file);
3469 sp.sp_mirror_id = (__u16)mirror_id;
3471 bias = MDS_CLOSE_LAYOUT_SPLIT;
3474 case LL_LEASE_PCC_ATTACH:
3475 if (ioc->lil_count != 1)
3478 arg += sizeof(*ioc);
3479 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3481 GOTO(out_lease_close, rc2 = -EFAULT);
3483 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3485 GOTO(out_lease_close, rc2);
3488 /* Grab latest data version */
3489 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3492 GOTO(out_lease_close, rc2);
3495 bias = MDS_PCC_ATTACH;
3498 /* without close intent */
3503 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3507 rc = ll_lease_och_release(inode, file);
3516 switch (ioc->lil_flags) {
3517 case LL_LEASE_RESYNC_DONE:
3519 OBD_FREE(data, data_size);
3521 case LL_LEASE_LAYOUT_MERGE:
3522 case LL_LEASE_LAYOUT_SPLIT:
3526 case LL_LEASE_PCC_ATTACH:
3529 rc = pcc_readwrite_attach_fini(file, inode,
3530 param.pa_layout_gen,
3537 rc = ll_lease_type_from_fmode(fmode);
3541 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3544 struct inode *inode = file_inode(file);
3545 struct ll_inode_info *lli = ll_i2info(inode);
3546 struct ll_file_data *fd = file->private_data;
3547 struct obd_client_handle *och = NULL;
3548 __u64 open_flags = 0;
3554 switch (ioc->lil_mode) {
3555 case LL_LEASE_WRLCK:
3556 if (!(file->f_mode & FMODE_WRITE))
3558 fmode = FMODE_WRITE;
3560 case LL_LEASE_RDLCK:
3561 if (!(file->f_mode & FMODE_READ))
3565 case LL_LEASE_UNLCK:
3566 RETURN(ll_file_unlock_lease(file, ioc, arg));
3571 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3573 /* apply for lease */
3574 if (ioc->lil_flags & LL_LEASE_RESYNC)
3575 open_flags = MDS_OPEN_RESYNC;
3576 och = ll_lease_open(inode, file, fmode, open_flags);
3578 RETURN(PTR_ERR(och));
3580 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3581 rc = ll_lease_file_resync(och, inode, arg);
3583 ll_lease_close(och, inode, NULL);
3586 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3588 ll_lease_close(och, inode, NULL);
3594 mutex_lock(&lli->lli_och_mutex);
3595 if (fd->fd_lease_och == NULL) {
3596 fd->fd_lease_och = och;
3599 mutex_unlock(&lli->lli_och_mutex);
3601 /* impossible now that only excl is supported for now */
3602 ll_lease_close(och, inode, &lease_broken);
3608 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3610 struct ll_inode_info *lli = ll_i2info(inode);
3611 struct ll_sb_info *sbi = ll_i2sbi(inode);
3612 __u64 now = ktime_get_real_seconds();
3615 spin_lock(&lli->lli_heat_lock);
3616 heat->lh_flags = lli->lli_heat_flags;
3617 for (i = 0; i < heat->lh_count; i++)
3618 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3619 now, sbi->ll_heat_decay_weight,
3620 sbi->ll_heat_period_second);
3621 spin_unlock(&lli->lli_heat_lock);
3624 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3626 struct ll_inode_info *lli = ll_i2info(inode);
3629 spin_lock(&lli->lli_heat_lock);
3630 if (flags & LU_HEAT_FLAG_CLEAR)
3631 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3633 if (flags & LU_HEAT_FLAG_OFF)
3634 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3636 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3638 spin_unlock(&lli->lli_heat_lock);
3644 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3646 struct inode *inode = file_inode(file);
3647 struct ll_file_data *fd = file->private_data;
3651 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3652 PFID(ll_inode2fid(inode)), inode, cmd);
3653 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3655 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3656 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3660 case LL_IOC_GETFLAGS:
3661 /* Get the current value of the file flags */
3662 return put_user(fd->fd_flags, (int __user *)arg);
3663 case LL_IOC_SETFLAGS:
3664 case LL_IOC_CLRFLAGS:
3665 /* Set or clear specific file flags */
3666 /* XXX This probably needs checks to ensure the flags are
3667 * not abused, and to handle any flag side effects.
3669 if (get_user(flags, (int __user *) arg))
3672 if (cmd == LL_IOC_SETFLAGS) {
3673 if ((flags & LL_FILE_IGNORE_LOCK) &&
3674 !(file->f_flags & O_DIRECT)) {
3675 CERROR("%s: unable to disable locking on "
3676 "non-O_DIRECT file\n", current->comm);
3680 fd->fd_flags |= flags;
3682 fd->fd_flags &= ~flags;
3685 case LL_IOC_LOV_SETSTRIPE:
3686 case LL_IOC_LOV_SETSTRIPE_NEW:
3687 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3688 case LL_IOC_LOV_SETEA:
3689 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3690 case LL_IOC_LOV_SWAP_LAYOUTS: {
3692 struct lustre_swap_layouts lsl;
3694 if (copy_from_user(&lsl, (char __user *)arg,
3695 sizeof(struct lustre_swap_layouts)))
3698 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3701 file2 = fget(lsl.sl_fd);
3705 /* O_WRONLY or O_RDWR */
3706 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3707 GOTO(out, rc = -EPERM);
3709 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3710 struct inode *inode2;
3711 struct ll_inode_info *lli;
3712 struct obd_client_handle *och = NULL;
3714 lli = ll_i2info(inode);
3715 mutex_lock(&lli->lli_och_mutex);
3716 if (fd->fd_lease_och != NULL) {
3717 och = fd->fd_lease_och;
3718 fd->fd_lease_och = NULL;
3720 mutex_unlock(&lli->lli_och_mutex);
3722 GOTO(out, rc = -ENOLCK);
3723 inode2 = file_inode(file2);
3724 rc = ll_swap_layouts_close(och, inode, inode2);
3726 rc = ll_swap_layouts(file, file2, &lsl);
3732 case LL_IOC_LOV_GETSTRIPE:
3733 case LL_IOC_LOV_GETSTRIPE_NEW:
3734 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
3735 case FS_IOC_GETFLAGS:
3736 case FS_IOC_SETFLAGS:
3737 RETURN(ll_iocontrol(inode, file, cmd, arg));
3738 case FSFILT_IOC_GETVERSION:
3739 case FS_IOC_GETVERSION:
3740 RETURN(put_user(inode->i_generation, (int __user *)arg));
3741 /* We need to special case any other ioctls we want to handle,
3742 * to send them to the MDS/OST as appropriate and to properly
3743 * network encode the arg field. */
3744 case FS_IOC_SETVERSION:
3747 case LL_IOC_GROUP_LOCK:
3748 RETURN(ll_get_grouplock(inode, file, arg));
3749 case LL_IOC_GROUP_UNLOCK:
3750 RETURN(ll_put_grouplock(inode, file, arg));
3751 case IOC_OBD_STATFS:
3752 RETURN(ll_obd_statfs(inode, (void __user *)arg));
3754 case LL_IOC_FLUSHCTX:
3755 RETURN(ll_flush_ctx(inode));
3756 case LL_IOC_PATH2FID: {
3757 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
3758 sizeof(struct lu_fid)))
3763 case LL_IOC_GETPARENT:
3764 RETURN(ll_getparent(file, (struct getparent __user *)arg));
3766 case OBD_IOC_FID2PATH:
3767 RETURN(ll_fid2path(inode, (void __user *)arg));
3768 case LL_IOC_DATA_VERSION: {
3769 struct ioc_data_version idv;
3772 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
3775 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
3776 rc = ll_ioc_data_version(inode, &idv);
3779 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
3785 case LL_IOC_GET_MDTIDX: {
3788 mdtidx = ll_get_mdt_idx(inode);
3792 if (put_user((int)mdtidx, (int __user *)arg))
3797 case OBD_IOC_GETDTNAME:
3798 case OBD_IOC_GETMDNAME:
3799 RETURN(ll_get_obd_name(inode, cmd, arg));
3800 case LL_IOC_HSM_STATE_GET: {
3801 struct md_op_data *op_data;
3802 struct hsm_user_state *hus;
3809 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3810 LUSTRE_OPC_ANY, hus);
3811 if (IS_ERR(op_data)) {
3813 RETURN(PTR_ERR(op_data));
3816 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3819 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
3822 ll_finish_md_op_data(op_data);
3826 case LL_IOC_HSM_STATE_SET: {
3827 struct hsm_state_set *hss;
3834 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
3839 rc = ll_hsm_state_set(inode, hss);
3844 case LL_IOC_HSM_ACTION: {
3845 struct md_op_data *op_data;
3846 struct hsm_current_action *hca;
3853 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3854 LUSTRE_OPC_ANY, hca);
3855 if (IS_ERR(op_data)) {
3857 RETURN(PTR_ERR(op_data));
3860 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
3863 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
3866 ll_finish_md_op_data(op_data);
3870 case LL_IOC_SET_LEASE_OLD: {
3871 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
3873 RETURN(ll_file_set_lease(file, &ioc, 0));
3875 case LL_IOC_SET_LEASE: {
3876 struct ll_ioc_lease ioc;
3878 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
3881 RETURN(ll_file_set_lease(file, &ioc, arg));
3883 case LL_IOC_GET_LEASE: {
3884 struct ll_inode_info *lli = ll_i2info(inode);
3885 struct ldlm_lock *lock = NULL;
3888 mutex_lock(&lli->lli_och_mutex);
3889 if (fd->fd_lease_och != NULL) {
3890 struct obd_client_handle *och = fd->fd_lease_och;
3892 lock = ldlm_handle2lock(&och->och_lease_handle);
3894 lock_res_and_lock(lock);
3895 if (!ldlm_is_cancel(lock))
3896 fmode = och->och_flags;
3898 unlock_res_and_lock(lock);
3899 LDLM_LOCK_PUT(lock);
3902 mutex_unlock(&lli->lli_och_mutex);
3904 RETURN(ll_lease_type_from_fmode(fmode));
3906 case LL_IOC_HSM_IMPORT: {
3907 struct hsm_user_import *hui;
3913 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
3918 rc = ll_hsm_import(inode, file, hui);
3923 case LL_IOC_FUTIMES_3: {
3924 struct ll_futimes_3 lfu;
3926 if (copy_from_user(&lfu,
3927 (const struct ll_futimes_3 __user *)arg,
3931 RETURN(ll_file_futimes_3(file, &lfu));
3933 case LL_IOC_LADVISE: {
3934 struct llapi_ladvise_hdr *k_ladvise_hdr;
3935 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
3938 int alloc_size = sizeof(*k_ladvise_hdr);
3941 u_ladvise_hdr = (void __user *)arg;
3942 OBD_ALLOC_PTR(k_ladvise_hdr);
3943 if (k_ladvise_hdr == NULL)
3946 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3947 GOTO(out_ladvise, rc = -EFAULT);
3949 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
3950 k_ladvise_hdr->lah_count < 1)
3951 GOTO(out_ladvise, rc = -EINVAL);
3953 num_advise = k_ladvise_hdr->lah_count;
3954 if (num_advise >= LAH_COUNT_MAX)
3955 GOTO(out_ladvise, rc = -EFBIG);
3957 OBD_FREE_PTR(k_ladvise_hdr);
3958 alloc_size = offsetof(typeof(*k_ladvise_hdr),
3959 lah_advise[num_advise]);
3960 OBD_ALLOC(k_ladvise_hdr, alloc_size);
3961 if (k_ladvise_hdr == NULL)
3965 * TODO: submit multiple advices to one server in a single RPC
3967 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
3968 GOTO(out_ladvise, rc = -EFAULT);
3970 for (i = 0; i < num_advise; i++) {
3971 struct llapi_lu_ladvise *k_ladvise =
3972 &k_ladvise_hdr->lah_advise[i];
3973 struct llapi_lu_ladvise __user *u_ladvise =
3974 &u_ladvise_hdr->lah_advise[i];
3976 rc = ll_ladvise_sanity(inode, k_ladvise);
3978 GOTO(out_ladvise, rc);
3980 switch (k_ladvise->lla_advice) {
3981 case LU_LADVISE_LOCKNOEXPAND:
3982 rc = ll_lock_noexpand(file,
3983 k_ladvise->lla_peradvice_flags);
3984 GOTO(out_ladvise, rc);
3985 case LU_LADVISE_LOCKAHEAD:
3987 rc = ll_file_lock_ahead(file, k_ladvise);
3990 GOTO(out_ladvise, rc);
3993 &u_ladvise->lla_lockahead_result))
3994 GOTO(out_ladvise, rc = -EFAULT);
3997 rc = ll_ladvise(inode, file,
3998 k_ladvise_hdr->lah_flags,
4001 GOTO(out_ladvise, rc);
4008 OBD_FREE(k_ladvise_hdr, alloc_size);
4011 case LL_IOC_FLR_SET_MIRROR: {
4012 /* mirror I/O must be direct to avoid polluting page cache
4014 if (!(file->f_flags & O_DIRECT))
4017 fd->fd_designated_mirror = (__u32)arg;
4020 case FS_IOC_FSGETXATTR:
4021 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
4022 case FS_IOC_FSSETXATTR:
4023 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
4025 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
4026 case LL_IOC_HEAT_GET: {
4027 struct lu_heat uheat;
4028 struct lu_heat *heat;
4031 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
4034 if (uheat.lh_count > OBD_HEAT_COUNT)
4035 uheat.lh_count = OBD_HEAT_COUNT;
4037 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4038 OBD_ALLOC(heat, size);
4042 heat->lh_count = uheat.lh_count;
4043 ll_heat_get(inode, heat);
4044 rc = copy_to_user((char __user *)arg, heat, size);
4045 OBD_FREE(heat, size);
4046 RETURN(rc ? -EFAULT : 0);
4048 case LL_IOC_HEAT_SET: {
4051 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
4054 rc = ll_heat_set(inode, flags);
4057 case LL_IOC_PCC_DETACH: {
4058 struct lu_pcc_detach *detach;
4060 OBD_ALLOC_PTR(detach);
4064 if (copy_from_user(detach,
4065 (const struct lu_pcc_detach __user *)arg,
4067 GOTO(out_detach_free, rc = -EFAULT);
4069 if (!S_ISREG(inode->i_mode))
4070 GOTO(out_detach_free, rc = -EINVAL);
4072 if (!inode_owner_or_capable(inode))
4073 GOTO(out_detach_free, rc = -EPERM);
4075 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4077 OBD_FREE_PTR(detach);
4080 case LL_IOC_PCC_STATE: {
4081 struct lu_pcc_state __user *ustate =
4082 (struct lu_pcc_state __user *)arg;
4083 struct lu_pcc_state *state;
4085 OBD_ALLOC_PTR(state);
4089 if (copy_from_user(state, ustate, sizeof(*state)))
4090 GOTO(out_state, rc = -EFAULT);
4092 rc = pcc_ioctl_state(file, inode, state);
4094 GOTO(out_state, rc);
4096 if (copy_to_user(ustate, state, sizeof(*state)))
4097 GOTO(out_state, rc = -EFAULT);
4100 OBD_FREE_PTR(state);
4103 #ifdef HAVE_LUSTRE_CRYPTO
4104 case LL_IOC_SET_ENCRYPTION_POLICY:
4105 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4107 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
4108 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
4109 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4111 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
4112 case LL_IOC_ADD_ENCRYPTION_KEY:
4113 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4115 return llcrypt_ioctl_add_key(file, (void __user *)arg);
4116 case LL_IOC_REMOVE_ENCRYPTION_KEY:
4117 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4119 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
4120 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
4121 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4123 return llcrypt_ioctl_remove_key_all_users(file,
4124 (void __user *)arg);
4125 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
4126 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4128 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
4131 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
4132 (void __user *)arg));
4136 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4138 struct inode *inode = file_inode(file);
4139 loff_t retval, eof = 0;
4140 ktime_t kstart = ktime_get();
4143 retval = offset + ((origin == SEEK_END) ? i_size_read(inode) :
4144 (origin == SEEK_CUR) ? file->f_pos : 0);
4145 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4146 PFID(ll_inode2fid(inode)), inode, retval, retval,
4149 if (origin == SEEK_END || origin == SEEK_HOLE || origin == SEEK_DATA) {
4150 retval = ll_glimpse_size(inode);
4153 eof = i_size_read(inode);
4156 retval = generic_file_llseek_size(file, offset, origin,
4157 ll_file_maxbytes(inode), eof);
4159 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4160 ktime_us_delta(ktime_get(), kstart));
4164 static int ll_flush(struct file *file, fl_owner_t id)
4166 struct inode *inode = file_inode(file);
4167 struct ll_inode_info *lli = ll_i2info(inode);
4168 struct ll_file_data *fd = file->private_data;
4171 LASSERT(!S_ISDIR(inode->i_mode));
4173 /* catch async errors that were recorded back when async writeback
4174 * failed for pages in this mapping. */
4175 rc = lli->lli_async_rc;
4176 lli->lli_async_rc = 0;
4177 if (lli->lli_clob != NULL) {
4178 err = lov_read_and_clear_async_rc(lli->lli_clob);
4183 /* The application has been told write failure already.
4184 * Do not report failure again. */
4185 if (fd->fd_write_failed)
4187 return rc ? -EIO : 0;
4191 * Called to make sure a portion of file has been written out.
4192 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4194 * Return how many pages have been written.
4196 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4197 enum cl_fsync_mode mode, int ignore_layout)
4201 struct cl_fsync_io *fio;
4206 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4207 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4210 env = cl_env_get(&refcheck);
4212 RETURN(PTR_ERR(env));
4214 io = vvp_env_thread_io(env);
4215 io->ci_obj = ll_i2info(inode)->lli_clob;
4216 io->ci_ignore_layout = ignore_layout;
4218 /* initialize parameters for sync */
4219 fio = &io->u.ci_fsync;
4220 fio->fi_start = start;
4222 fio->fi_fid = ll_inode2fid(inode);
4223 fio->fi_mode = mode;
4224 fio->fi_nr_written = 0;
4226 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4227 result = cl_io_loop(env, io);
4229 result = io->ci_result;
4231 result = fio->fi_nr_written;
4232 cl_io_fini(env, io);
4233 cl_env_put(env, &refcheck);
4239 * When dentry is provided (the 'else' case), file_dentry() may be
4240 * null and dentry must be used directly rather than pulled from
4241 * file_dentry() as is done otherwise.
4244 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4246 struct dentry *dentry = file_dentry(file);
4247 struct inode *inode = dentry->d_inode;
4248 struct ll_inode_info *lli = ll_i2info(inode);
4249 struct ptlrpc_request *req;
4250 ktime_t kstart = ktime_get();
4256 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4257 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4259 /* fsync's caller has already called _fdata{sync,write}, we want
4260 * that IO to finish before calling the osc and mdc sync methods */
4261 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4264 /* catch async errors that were recorded back when async writeback
4265 * failed for pages in this mapping. */
4266 if (!S_ISDIR(inode->i_mode)) {
4267 err = lli->lli_async_rc;
4268 lli->lli_async_rc = 0;
4271 if (lli->lli_clob != NULL) {
4272 err = lov_read_and_clear_async_rc(lli->lli_clob);
4278 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4282 ptlrpc_req_finished(req);
4284 if (S_ISREG(inode->i_mode)) {
4285 struct ll_file_data *fd = file->private_data;
4288 /* Sync metadata on MDT first, and then sync the cached data
4291 err = pcc_fsync(file, start, end, datasync, &cached);
4293 err = cl_sync_file_range(inode, start, end,
4295 if (rc == 0 && err < 0)
4298 fd->fd_write_failed = true;
4300 fd->fd_write_failed = false;
4303 inode_unlock(inode);
4306 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4307 ktime_us_delta(ktime_get(), kstart));
4312 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4314 struct inode *inode = file_inode(file);
4315 struct ll_sb_info *sbi = ll_i2sbi(inode);
4316 struct ldlm_enqueue_info einfo = {
4317 .ei_type = LDLM_FLOCK,
4318 .ei_cb_cp = ldlm_flock_completion_ast,
4319 .ei_cbdata = file_lock,
4321 struct md_op_data *op_data;
4322 struct lustre_handle lockh = { 0 };
4323 union ldlm_policy_data flock = { { 0 } };
4324 int fl_type = file_lock->fl_type;
4325 ktime_t kstart = ktime_get();
4331 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4332 PFID(ll_inode2fid(inode)), file_lock);
4334 if (file_lock->fl_flags & FL_FLOCK) {
4335 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4336 /* flocks are whole-file locks */
4337 flock.l_flock.end = OFFSET_MAX;
4338 /* For flocks owner is determined by the local file desctiptor*/
4339 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4340 } else if (file_lock->fl_flags & FL_POSIX) {
4341 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4342 flock.l_flock.start = file_lock->fl_start;
4343 flock.l_flock.end = file_lock->fl_end;
4347 flock.l_flock.pid = file_lock->fl_pid;
4349 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4350 /* Somewhat ugly workaround for svc lockd.
4351 * lockd installs custom fl_lmops->lm_compare_owner that checks
4352 * for the fl_owner to be the same (which it always is on local node
4353 * I guess between lockd processes) and then compares pid.
4354 * As such we assign pid to the owner field to make it all work,
4355 * conflict with normal locks is unlikely since pid space and
4356 * pointer space for current->files are not intersecting */
4357 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4358 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4363 einfo.ei_mode = LCK_PR;
4366 /* An unlock request may or may not have any relation to
4367 * existing locks so we may not be able to pass a lock handle
4368 * via a normal ldlm_lock_cancel() request. The request may even
4369 * unlock a byte range in the middle of an existing lock. In
4370 * order to process an unlock request we need all of the same
4371 * information that is given with a normal read or write record
4372 * lock request. To avoid creating another ldlm unlock (cancel)
4373 * message we'll treat a LCK_NL flock request as an unlock. */
4374 einfo.ei_mode = LCK_NL;
4377 einfo.ei_mode = LCK_PW;
4380 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4395 flags = LDLM_FL_BLOCK_NOWAIT;
4401 flags = LDLM_FL_TEST_LOCK;
4404 CERROR("unknown fcntl lock command: %d\n", cmd);
4408 /* Save the old mode so that if the mode in the lock changes we
4409 * can decrement the appropriate reader or writer refcount. */
4410 file_lock->fl_type = einfo.ei_mode;
4412 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4413 LUSTRE_OPC_ANY, NULL);
4414 if (IS_ERR(op_data))
4415 RETURN(PTR_ERR(op_data));
4417 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4418 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4419 flock.l_flock.pid, flags, einfo.ei_mode,
4420 flock.l_flock.start, flock.l_flock.end);
4422 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4425 /* Restore the file lock type if not TEST lock. */
4426 if (!(flags & LDLM_FL_TEST_LOCK))
4427 file_lock->fl_type = fl_type;
4429 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4430 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4431 !(flags & LDLM_FL_TEST_LOCK))
4432 rc2 = locks_lock_file_wait(file, file_lock);
4434 if ((file_lock->fl_flags & FL_FLOCK) &&
4435 (rc == 0 || file_lock->fl_type == F_UNLCK))
4436 rc2 = flock_lock_file_wait(file, file_lock);
4437 if ((file_lock->fl_flags & FL_POSIX) &&
4438 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4439 !(flags & LDLM_FL_TEST_LOCK))
4440 rc2 = posix_lock_file_wait(file, file_lock);
4441 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4443 if (rc2 && file_lock->fl_type != F_UNLCK) {
4444 einfo.ei_mode = LCK_NL;
4445 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4450 ll_finish_md_op_data(op_data);
4453 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
4454 ktime_us_delta(ktime_get(), kstart));
4458 int ll_get_fid_by_name(struct inode *parent, const char *name,
4459 int namelen, struct lu_fid *fid,
4460 struct inode **inode)
4462 struct md_op_data *op_data = NULL;
4463 struct mdt_body *body;
4464 struct ptlrpc_request *req;
4468 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4469 LUSTRE_OPC_ANY, NULL);
4470 if (IS_ERR(op_data))
4471 RETURN(PTR_ERR(op_data));
4473 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4474 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4475 ll_finish_md_op_data(op_data);
4479 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4481 GOTO(out_req, rc = -EFAULT);
4483 *fid = body->mbo_fid1;
4486 rc = ll_prep_inode(inode, req, parent->i_sb, NULL);
4488 ptlrpc_req_finished(req);
4492 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4495 struct dentry *dchild = NULL;
4496 struct inode *child_inode = NULL;
4497 struct md_op_data *op_data;
4498 struct ptlrpc_request *request = NULL;
4499 struct obd_client_handle *och = NULL;
4501 struct mdt_body *body;
4502 __u64 data_version = 0;
4503 size_t namelen = strlen(name);
4504 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4508 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4509 PFID(ll_inode2fid(parent)), name,
4510 lum->lum_stripe_offset, lum->lum_stripe_count);
4512 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4513 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4514 lustre_swab_lmv_user_md(lum);
4516 /* Get child FID first */
4517 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4520 dchild = d_lookup(file_dentry(file), &qstr);
4522 if (dchild->d_inode)
4523 child_inode = igrab(dchild->d_inode);
4528 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4537 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4538 OBD_CONNECT2_DIR_MIGRATE)) {
4539 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4540 ll_dir_striped(child_inode)) {
4541 CERROR("%s: MDT doesn't support stripe directory "
4542 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4543 GOTO(out_iput, rc = -EOPNOTSUPP);
4548 * lfs migrate command needs to be blocked on the client
4549 * by checking the migrate FID against the FID of the
4552 if (child_inode == parent->i_sb->s_root->d_inode)
4553 GOTO(out_iput, rc = -EINVAL);
4555 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4556 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
4557 if (IS_ERR(op_data))
4558 GOTO(out_iput, rc = PTR_ERR(op_data));
4560 inode_lock(child_inode);
4561 op_data->op_fid3 = *ll_inode2fid(child_inode);
4562 if (!fid_is_sane(&op_data->op_fid3)) {
4563 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4564 ll_i2sbi(parent)->ll_fsname, name,
4565 PFID(&op_data->op_fid3));
4566 GOTO(out_unlock, rc = -EINVAL);
4569 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4570 op_data->op_data = lum;
4571 op_data->op_data_size = lumlen;
4574 if (S_ISREG(child_inode->i_mode)) {
4575 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4579 GOTO(out_unlock, rc);
4582 rc = ll_data_version(child_inode, &data_version,
4585 GOTO(out_close, rc);
4587 op_data->op_open_handle = och->och_open_handle;
4588 op_data->op_data_version = data_version;
4589 op_data->op_lease_handle = och->och_lease_handle;
4590 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4592 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4593 och->och_mod->mod_open_req->rq_replay = 0;
4594 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4597 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data, name, namelen,
4598 name, namelen, &request);
4600 LASSERT(request != NULL);
4601 ll_update_times(request, parent);
4604 if (rc == 0 || rc == -EAGAIN) {
4605 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4606 LASSERT(body != NULL);
4608 /* If the server does release layout lock, then we cleanup
4609 * the client och here, otherwise release it in out_close: */
4610 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4611 obd_mod_put(och->och_mod);
4612 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4614 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4620 if (request != NULL) {
4621 ptlrpc_req_finished(request);
4625 /* Try again if the lease has cancelled. */
4626 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4631 ll_lease_close(och, child_inode, NULL);
4633 clear_nlink(child_inode);
4635 inode_unlock(child_inode);
4636 ll_finish_md_op_data(op_data);
4643 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
4645 struct ll_file_data *fd = file->private_data;
4649 * In order to avoid flood of warning messages, only print one message
4650 * for one file. And the entire message rate on the client is limited
4651 * by CDEBUG_LIMIT too.
4653 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
4654 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
4655 CDEBUG_LIMIT(D_TTY | D_CONSOLE,
4656 "flock disabled, mount with '-o [local]flock' to enable\r\n");
4662 * test if some locks matching bits and l_req_mode are acquired
4663 * - bits can be in different locks
4664 * - if found clear the common lock bits in *bits
4665 * - the bits not found, are kept in *bits
4667 * \param bits [IN] searched lock bits [IN]
4668 * \param l_req_mode [IN] searched lock mode
4669 * \retval boolean, true iff all bits are found
4671 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
4673 struct lustre_handle lockh;
4674 union ldlm_policy_data policy;
4675 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
4676 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
4685 fid = &ll_i2info(inode)->lli_fid;
4686 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
4687 ldlm_lockname[mode]);
4689 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
4690 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
4691 policy.l_inodebits.bits = *bits & BIT(i);
4692 if (policy.l_inodebits.bits == 0)
4695 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
4696 &policy, mode, &lockh)) {
4697 struct ldlm_lock *lock;
4699 lock = ldlm_handle2lock(&lockh);
4702 ~(lock->l_policy_data.l_inodebits.bits);
4703 LDLM_LOCK_PUT(lock);
4705 *bits &= ~policy.l_inodebits.bits;
4712 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
4713 struct lustre_handle *lockh, __u64 flags,
4714 enum ldlm_mode mode)
4716 union ldlm_policy_data policy = { .l_inodebits = { bits } };
4721 fid = &ll_i2info(inode)->lli_fid;
4722 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
4724 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
4725 fid, LDLM_IBITS, &policy, mode, lockh);
4730 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
4732 /* Already unlinked. Just update nlink and return success */
4733 if (rc == -ENOENT) {
4735 /* If it is striped directory, and there is bad stripe
4736 * Let's revalidate the dentry again, instead of returning
4738 if (ll_dir_striped(inode))
4741 /* This path cannot be hit for regular files unless in
4742 * case of obscure races, so no need to to validate
4744 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
4746 } else if (rc != 0) {
4747 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
4748 "%s: revalidate FID "DFID" error: rc = %d\n",
4749 ll_i2sbi(inode)->ll_fsname,
4750 PFID(ll_inode2fid(inode)), rc);
4756 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
4758 struct inode *parent;
4759 struct inode *inode = dentry->d_inode;
4760 struct obd_export *exp = ll_i2mdexp(inode);
4761 struct lookup_intent oit = {
4764 struct ptlrpc_request *req = NULL;
4765 struct md_op_data *op_data;
4769 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
4770 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
4772 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
4773 parent = dentry->d_parent->d_inode;
4777 /* Call getattr by fid, so do not provide name at all. */
4778 op_data = ll_prep_md_op_data(NULL, parent, inode, NULL, 0, 0,
4779 LUSTRE_OPC_ANY, NULL);
4780 if (IS_ERR(op_data))
4781 RETURN(PTR_ERR(op_data));
4783 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
4784 ll_finish_md_op_data(op_data);
4786 rc = ll_inode_revalidate_fini(inode, rc);
4790 rc = ll_revalidate_it_finish(req, &oit, dentry);
4792 ll_intent_release(&oit);
4796 /* Unlinked? Unhash dentry, so it is not picked up later by
4797 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
4798 * here to preserve get_cwd functionality on 2.6.
4800 if (!dentry->d_inode->i_nlink)
4801 d_lustre_invalidate(dentry);
4803 ll_lookup_finish_locks(&oit, dentry);
4805 ptlrpc_req_finished(req);
4810 static int ll_merge_md_attr(struct inode *inode)
4812 struct ll_inode_info *lli = ll_i2info(inode);
4813 struct cl_attr attr = { 0 };
4816 LASSERT(lli->lli_lsm_md != NULL);
4818 if (!lmv_dir_striped(lli->lli_lsm_md))
4821 down_read(&lli->lli_lsm_sem);
4822 rc = md_merge_attr(ll_i2mdexp(inode), &lli->lli_fid, lli->lli_lsm_md,
4823 &attr, ll_md_blocking_ast);
4824 up_read(&lli->lli_lsm_sem);
4828 set_nlink(inode, attr.cat_nlink);
4829 inode->i_blocks = attr.cat_blocks;
4830 i_size_write(inode, attr.cat_size);
4832 ll_i2info(inode)->lli_atime = attr.cat_atime;
4833 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
4834 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
4839 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
4842 struct inode *inode = de->d_inode;
4843 struct ll_sb_info *sbi = ll_i2sbi(inode);
4844 struct ll_inode_info *lli = ll_i2info(inode);
4845 struct inode *dir = de->d_parent->d_inode;
4846 bool need_glimpse = true;
4847 ktime_t kstart = ktime_get();
4850 /* The OST object(s) determine the file size, blocks and mtime. */
4851 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
4852 request_mask & STATX_MTIME))
4853 need_glimpse = false;
4855 if (dentry_may_statahead(dir, de))
4856 ll_start_statahead(dir, de, need_glimpse &&
4857 !(flags & AT_STATX_DONT_SYNC));
4859 if (flags & AT_STATX_DONT_SYNC)
4860 GOTO(fill_attr, rc = 0);
4862 rc = ll_inode_revalidate(de, IT_GETATTR);
4866 if (S_ISREG(inode->i_mode)) {
4870 GOTO(fill_attr, rc);
4872 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
4873 if (cached && rc < 0)
4877 GOTO(fill_attr, rc);
4880 * If the returned attr is masked with OBD_MD_FLSIZE &
4881 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
4882 * or blocks obtained from MDT is strictly correct, and the file
4883 * is usually not being modified by clients, and the [a|m|c]time
4884 * got from MDT is also strictly correct.
4885 * Under this circumstance, it does not need to send glimpse
4886 * RPCs to OSTs for file attributes such as the size and blocks.
4888 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
4889 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
4890 lli->lli_attr_valid & OBD_MD_FLMTIME) {
4891 inode->i_mtime.tv_sec = lli->lli_mtime;
4892 if (lli->lli_attr_valid & OBD_MD_FLATIME)
4893 inode->i_atime.tv_sec = lli->lli_atime;
4894 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
4895 inode->i_ctime.tv_sec = lli->lli_ctime;
4896 GOTO(fill_attr, rc);
4899 /* In case of restore, the MDT has the right size and has
4900 * already send it back without granting the layout lock,
4901 * inode is up-to-date so glimpse is useless.
4902 * Also to glimpse we need the layout, in case of a running
4903 * restore the MDT holds the layout lock so the glimpse will
4904 * block up to the end of restore (getattr will block)
4906 if (!ll_file_test_flag(lli, LLIF_FILE_RESTORING)) {
4907 rc = ll_glimpse_size(inode);
4912 /* If object isn't regular a file then don't validate size. */
4913 if (ll_dir_striped(inode)) {
4914 rc = ll_merge_md_attr(inode);
4919 if (lli->lli_attr_valid & OBD_MD_FLATIME)
4920 inode->i_atime.tv_sec = lli->lli_atime;
4921 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
4922 inode->i_mtime.tv_sec = lli->lli_mtime;
4923 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
4924 inode->i_ctime.tv_sec = lli->lli_ctime;
4928 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
4930 if (ll_need_32bit_api(sbi)) {
4931 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
4932 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
4933 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
4935 stat->ino = inode->i_ino;
4936 stat->dev = inode->i_sb->s_dev;
4937 stat->rdev = inode->i_rdev;
4940 stat->mode = inode->i_mode;
4941 stat->uid = inode->i_uid;
4942 stat->gid = inode->i_gid;
4943 stat->atime = inode->i_atime;
4944 stat->mtime = inode->i_mtime;
4945 stat->ctime = inode->i_ctime;
4946 /* stat->blksize is used to tell about preferred IO size */
4947 if (sbi->ll_stat_blksize)
4948 stat->blksize = sbi->ll_stat_blksize;
4949 else if (S_ISREG(inode->i_mode))
4950 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
4951 LL_MAX_BLKSIZE_BITS);
4953 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
4955 stat->nlink = inode->i_nlink;
4956 stat->size = i_size_read(inode);
4957 stat->blocks = inode->i_blocks;
4959 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
4960 if (flags & AT_STATX_DONT_SYNC) {
4961 if (stat->size == 0 &&
4962 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
4963 stat->size = lli->lli_lazysize;
4964 if (stat->blocks == 0 &&
4965 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
4966 stat->blocks = lli->lli_lazyblocks;
4969 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
4970 stat->result_mask |= STATX_BTIME;
4971 stat->btime.tv_sec = lli->lli_btime;
4974 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
4975 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
4976 stat->result_mask &= request_mask;
4979 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
4980 ktime_us_delta(ktime_get(), kstart));
4985 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
4986 int ll_getattr(const struct path *path, struct kstat *stat,
4987 u32 request_mask, unsigned int flags)
4989 return ll_getattr_dentry(path->dentry, stat, request_mask, flags);
4992 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
4994 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
4995 AT_STATX_SYNC_AS_STAT);
4999 int cl_falloc(struct inode *inode, int mode, loff_t offset, loff_t len)
5004 int rc; loff_t sa_falloc_end;
5005 loff_t size = i_size_read(inode);
5009 env = cl_env_get(&refcheck);
5011 RETURN(PTR_ERR(env));
5013 io = vvp_env_thread_io(env);
5014 io->ci_obj = ll_i2info(inode)->lli_clob;
5015 io->ci_verify_layout = 1;
5016 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5017 io->u.ci_setattr.sa_falloc_mode = mode;
5018 io->u.ci_setattr.sa_falloc_offset = offset;
5019 io->u.ci_setattr.sa_falloc_len = len;
5020 io->u.ci_setattr.sa_falloc_end = io->u.ci_setattr.sa_falloc_offset +
5021 io->u.ci_setattr.sa_falloc_len;
5022 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5023 sa_falloc_end = io->u.ci_setattr.sa_falloc_end;
5024 if (sa_falloc_end > size) {
5025 /* Check new size against VFS/VM file size limit and rlimit */
5026 rc = inode_newsize_ok(inode, sa_falloc_end);
5029 if (sa_falloc_end > ll_file_maxbytes(inode)) {
5030 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5031 (unsigned long long)(sa_falloc_end),
5032 ll_file_maxbytes(inode));
5036 io->u.ci_setattr.sa_attr.lvb_size = sa_falloc_end;
5037 if (!(mode & FALLOC_FL_KEEP_SIZE))
5038 io->u.ci_setattr.sa_avalid |= ATTR_SIZE;
5040 io->u.ci_setattr.sa_attr.lvb_size = size;
5044 if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0)
5045 rc = cl_io_loop(env, io);
5049 cl_io_fini(env, io);
5050 if (unlikely(io->ci_need_restart))
5054 cl_env_put(env, &refcheck);
5058 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5060 struct inode *inode = filp->f_path.dentry->d_inode;
5064 * Encrypted inodes can't handle collapse range or zero range or insert
5065 * range since we would need to re-encrypt blocks with a different IV or
5066 * XTS tweak (which are based on the logical block number).
5067 * Similar to what ext4 does.
5069 if (IS_ENCRYPTED(inode) &&
5070 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5071 FALLOC_FL_ZERO_RANGE)))
5072 RETURN(-EOPNOTSUPP);
5075 * Only mode == 0 (which is standard prealloc) is supported now.
5076 * Punch is not supported yet.
5078 if (mode & ~FALLOC_FL_KEEP_SIZE)
5079 RETURN(-EOPNOTSUPP);
5081 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5083 rc = cl_falloc(inode, mode, offset, len);
5088 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5089 __u64 start, __u64 len)
5093 struct fiemap *fiemap;
5094 unsigned int extent_count = fieinfo->fi_extents_max;
5096 num_bytes = sizeof(*fiemap) + (extent_count *
5097 sizeof(struct fiemap_extent));
5098 OBD_ALLOC_LARGE(fiemap, num_bytes);
5103 fiemap->fm_flags = fieinfo->fi_flags;
5104 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5105 fiemap->fm_start = start;
5106 fiemap->fm_length = len;
5107 if (extent_count > 0 &&
5108 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5109 sizeof(struct fiemap_extent)) != 0)
5110 GOTO(out, rc = -EFAULT);
5112 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5114 fieinfo->fi_flags = fiemap->fm_flags;
5115 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5116 if (extent_count > 0 &&
5117 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5118 fiemap->fm_mapped_extents *
5119 sizeof(struct fiemap_extent)) != 0)
5120 GOTO(out, rc = -EFAULT);
5122 OBD_FREE_LARGE(fiemap, num_bytes);
5126 struct posix_acl *ll_get_acl(struct inode *inode, int type)
5128 struct ll_inode_info *lli = ll_i2info(inode);
5129 struct posix_acl *acl = NULL;
5132 spin_lock(&lli->lli_lock);
5133 /* VFS' acl_permission_check->check_acl will release the refcount */
5134 acl = posix_acl_dup(lli->lli_posix_acl);
5135 spin_unlock(&lli->lli_lock);
5140 #ifdef HAVE_IOP_SET_ACL
5141 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
5142 int ll_set_acl(struct inode *inode, struct posix_acl *acl, int type)
5144 struct ll_sb_info *sbi = ll_i2sbi(inode);
5145 struct ptlrpc_request *req = NULL;
5146 const char *name = NULL;
5148 size_t value_size = 0;
5153 case ACL_TYPE_ACCESS:
5154 name = XATTR_NAME_POSIX_ACL_ACCESS;
5156 rc = posix_acl_update_mode(inode, &inode->i_mode, &acl);
5159 case ACL_TYPE_DEFAULT:
5160 name = XATTR_NAME_POSIX_ACL_DEFAULT;
5161 if (!S_ISDIR(inode->i_mode))
5162 rc = acl ? -EACCES : 0;
5173 value_size = posix_acl_xattr_size(acl->a_count);
5174 value = kmalloc(value_size, GFP_NOFS);
5176 GOTO(out, rc = -ENOMEM);
5178 rc = posix_acl_to_xattr(&init_user_ns, acl, value, value_size);
5180 GOTO(out_value, rc);
5183 rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode),
5184 value ? OBD_MD_FLXATTR : OBD_MD_FLXATTRRM,
5185 name, value, value_size, 0, 0, &req);
5187 ptlrpc_req_finished(req);
5192 forget_cached_acl(inode, type);
5194 set_cached_acl(inode, type, acl);
5197 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
5198 #endif /* HAVE_IOP_SET_ACL */
5200 int ll_inode_permission(struct inode *inode, int mask)
5203 struct ll_sb_info *sbi;
5204 struct root_squash_info *squash;
5205 struct cred *cred = NULL;
5206 const struct cred *old_cred = NULL;
5208 bool squash_id = false;
5209 ktime_t kstart = ktime_get();
5213 if (mask & MAY_NOT_BLOCK)
5217 * as root inode are NOT getting validated in lookup operation,
5218 * need to do it before permission check.
5221 if (inode == inode->i_sb->s_root->d_inode) {
5222 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
5227 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5228 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5230 /* squash fsuid/fsgid if needed */
5231 sbi = ll_i2sbi(inode);
5232 squash = &sbi->ll_squash;
5233 if (unlikely(squash->rsi_uid != 0 &&
5234 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5235 !(sbi->ll_flags & LL_SBI_NOROOTSQUASH))) {
5239 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5240 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5241 squash->rsi_uid, squash->rsi_gid);
5243 /* update current process's credentials
5244 * and FS capability */
5245 cred = prepare_creds();
5249 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5250 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5251 for (cap = 0; cap < sizeof(cfs_cap_t) * 8; cap++) {
5252 if (BIT(cap) & CFS_CAP_FS_MASK)
5253 cap_lower(cred->cap_effective, cap);
5255 old_cred = override_creds(cred);
5258 rc = generic_permission(inode, mask);
5259 /* restore current process's credentials and FS capability */
5261 revert_creds(old_cred);
5266 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5267 ktime_us_delta(ktime_get(), kstart));
5272 /* -o localflock - only provides locally consistent flock locks */
5273 struct file_operations ll_file_operations = {
5274 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5275 # ifdef HAVE_SYNC_READ_WRITE
5276 .read = new_sync_read,
5277 .write = new_sync_write,
5279 .read_iter = ll_file_read_iter,
5280 .write_iter = ll_file_write_iter,
5281 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5282 .read = ll_file_read,
5283 .aio_read = ll_file_aio_read,
5284 .write = ll_file_write,
5285 .aio_write = ll_file_aio_write,
5286 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5287 .unlocked_ioctl = ll_file_ioctl,
5288 .open = ll_file_open,
5289 .release = ll_file_release,
5290 .mmap = ll_file_mmap,
5291 .llseek = ll_file_seek,
5292 .splice_read = ll_file_splice_read,
5295 .fallocate = ll_fallocate,
5298 struct file_operations ll_file_operations_flock = {
5299 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5300 # ifdef HAVE_SYNC_READ_WRITE
5301 .read = new_sync_read,
5302 .write = new_sync_write,
5303 # endif /* HAVE_SYNC_READ_WRITE */
5304 .read_iter = ll_file_read_iter,
5305 .write_iter = ll_file_write_iter,
5306 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5307 .read = ll_file_read,
5308 .aio_read = ll_file_aio_read,
5309 .write = ll_file_write,
5310 .aio_write = ll_file_aio_write,
5311 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5312 .unlocked_ioctl = ll_file_ioctl,
5313 .open = ll_file_open,
5314 .release = ll_file_release,
5315 .mmap = ll_file_mmap,
5316 .llseek = ll_file_seek,
5317 .splice_read = ll_file_splice_read,
5320 .flock = ll_file_flock,
5321 .lock = ll_file_flock,
5322 .fallocate = ll_fallocate,
5325 /* These are for -o noflock - to return ENOSYS on flock calls */
5326 struct file_operations ll_file_operations_noflock = {
5327 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5328 # ifdef HAVE_SYNC_READ_WRITE
5329 .read = new_sync_read,
5330 .write = new_sync_write,
5331 # endif /* HAVE_SYNC_READ_WRITE */
5332 .read_iter = ll_file_read_iter,
5333 .write_iter = ll_file_write_iter,
5334 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5335 .read = ll_file_read,
5336 .aio_read = ll_file_aio_read,
5337 .write = ll_file_write,
5338 .aio_write = ll_file_aio_write,
5339 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5340 .unlocked_ioctl = ll_file_ioctl,
5341 .open = ll_file_open,
5342 .release = ll_file_release,
5343 .mmap = ll_file_mmap,
5344 .llseek = ll_file_seek,
5345 .splice_read = ll_file_splice_read,
5348 .flock = ll_file_noflock,
5349 .lock = ll_file_noflock,
5350 .fallocate = ll_fallocate,
5353 struct inode_operations ll_file_inode_operations = {
5354 .setattr = ll_setattr,
5355 .getattr = ll_getattr,
5356 .permission = ll_inode_permission,
5357 #ifdef HAVE_IOP_XATTR
5358 .setxattr = ll_setxattr,
5359 .getxattr = ll_getxattr,
5360 .removexattr = ll_removexattr,
5362 .listxattr = ll_listxattr,
5363 .fiemap = ll_fiemap,
5364 .get_acl = ll_get_acl,
5365 #ifdef HAVE_IOP_SET_ACL
5366 .set_acl = ll_set_acl,
5370 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5372 struct ll_inode_info *lli = ll_i2info(inode);
5373 struct cl_object *obj = lli->lli_clob;
5382 env = cl_env_get(&refcheck);
5384 RETURN(PTR_ERR(env));
5386 rc = cl_conf_set(env, lli->lli_clob, conf);
5390 if (conf->coc_opc == OBJECT_CONF_SET) {
5391 struct ldlm_lock *lock = conf->coc_lock;
5392 struct cl_layout cl = {
5396 LASSERT(lock != NULL);
5397 LASSERT(ldlm_has_layout(lock));
5399 /* it can only be allowed to match after layout is
5400 * applied to inode otherwise false layout would be
5401 * seen. Applying layout shoud happen before dropping
5402 * the intent lock. */
5403 ldlm_lock_allow_match(lock);
5405 rc = cl_object_layout_get(env, obj, &cl);
5410 DFID": layout version change: %u -> %u\n",
5411 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5413 ll_layout_version_set(lli, cl.cl_layout_gen);
5417 cl_env_put(env, &refcheck);
5422 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5423 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5426 struct ll_sb_info *sbi = ll_i2sbi(inode);
5427 struct ptlrpc_request *req;
5434 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5435 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5436 lock->l_lvb_data, lock->l_lvb_len);
5438 if (lock->l_lvb_data != NULL)
5441 /* if layout lock was granted right away, the layout is returned
5442 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5443 * blocked and then granted via completion ast, we have to fetch
5444 * layout here. Please note that we can't use the LVB buffer in
5445 * completion AST because it doesn't have a large enough buffer */
5446 rc = ll_get_default_mdsize(sbi, &lmmsize);
5450 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5451 XATTR_NAME_LOV, lmmsize, &req);
5454 GOTO(out, rc = 0); /* empty layout */
5461 if (lmmsize == 0) /* empty layout */
5464 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5466 GOTO(out, rc = -EFAULT);
5468 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5469 if (lvbdata == NULL)
5470 GOTO(out, rc = -ENOMEM);
5472 memcpy(lvbdata, lmm, lmmsize);
5473 lock_res_and_lock(lock);
5474 if (unlikely(lock->l_lvb_data == NULL)) {
5475 lock->l_lvb_type = LVB_T_LAYOUT;
5476 lock->l_lvb_data = lvbdata;
5477 lock->l_lvb_len = lmmsize;
5480 unlock_res_and_lock(lock);
5483 OBD_FREE_LARGE(lvbdata, lmmsize);
5488 ptlrpc_req_finished(req);
5493 * Apply the layout to the inode. Layout lock is held and will be released
5496 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5497 struct inode *inode)
5499 struct ll_inode_info *lli = ll_i2info(inode);
5500 struct ll_sb_info *sbi = ll_i2sbi(inode);
5501 struct ldlm_lock *lock;
5502 struct cl_object_conf conf;
5505 bool wait_layout = false;
5508 LASSERT(lustre_handle_is_used(lockh));
5510 lock = ldlm_handle2lock(lockh);
5511 LASSERT(lock != NULL);
5512 LASSERT(ldlm_has_layout(lock));
5514 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5515 PFID(&lli->lli_fid), inode);
5517 /* in case this is a caching lock and reinstate with new inode */
5518 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5520 lock_res_and_lock(lock);
5521 lvb_ready = ldlm_is_lvb_ready(lock);
5522 unlock_res_and_lock(lock);
5524 /* checking lvb_ready is racy but this is okay. The worst case is
5525 * that multi processes may configure the file on the same time. */
5529 rc = ll_layout_fetch(inode, lock);
5533 /* for layout lock, lmm is stored in lock's lvb.
5534 * lvb_data is immutable if the lock is held so it's safe to access it
5537 * set layout to file. Unlikely this will fail as old layout was
5538 * surely eliminated */
5539 memset(&conf, 0, sizeof conf);
5540 conf.coc_opc = OBJECT_CONF_SET;
5541 conf.coc_inode = inode;
5542 conf.coc_lock = lock;
5543 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5544 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5545 rc = ll_layout_conf(inode, &conf);
5547 /* refresh layout failed, need to wait */
5548 wait_layout = rc == -EBUSY;
5551 LDLM_LOCK_PUT(lock);
5552 ldlm_lock_decref(lockh, mode);
5554 /* wait for IO to complete if it's still being used. */
5556 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5557 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5559 memset(&conf, 0, sizeof conf);
5560 conf.coc_opc = OBJECT_CONF_WAIT;
5561 conf.coc_inode = inode;
5562 rc = ll_layout_conf(inode, &conf);
5566 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5567 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5573 * Issue layout intent RPC to MDS.
5574 * \param inode [in] file inode
5575 * \param intent [in] layout intent
5577 * \retval 0 on success
5578 * \retval < 0 error code
5580 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5582 struct ll_inode_info *lli = ll_i2info(inode);
5583 struct ll_sb_info *sbi = ll_i2sbi(inode);
5584 struct md_op_data *op_data;
5585 struct lookup_intent it;
5586 struct ptlrpc_request *req;
5590 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5591 0, 0, LUSTRE_OPC_ANY, NULL);
5592 if (IS_ERR(op_data))
5593 RETURN(PTR_ERR(op_data));
5595 op_data->op_data = intent;
5596 op_data->op_data_size = sizeof(*intent);
5598 memset(&it, 0, sizeof(it));
5599 it.it_op = IT_LAYOUT;
5600 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5601 intent->li_opc == LAYOUT_INTENT_TRUNC)
5602 it.it_flags = FMODE_WRITE;
5604 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5605 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5607 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5608 &ll_md_blocking_ast, 0);
5609 if (it.it_request != NULL)
5610 ptlrpc_req_finished(it.it_request);
5611 it.it_request = NULL;
5613 ll_finish_md_op_data(op_data);
5615 /* set lock data in case this is a new lock */
5617 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5619 ll_intent_drop_lock(&it);
5625 * This function checks if there exists a LAYOUT lock on the client side,
5626 * or enqueues it if it doesn't have one in cache.
5628 * This function will not hold layout lock so it may be revoked any time after
5629 * this function returns. Any operations depend on layout should be redone
5632 * This function should be called before lov_io_init() to get an uptodate
5633 * layout version, the caller should save the version number and after IO
5634 * is finished, this function should be called again to verify that layout
5635 * is not changed during IO time.
5637 int ll_layout_refresh(struct inode *inode, __u32 *gen)
5639 struct ll_inode_info *lli = ll_i2info(inode);
5640 struct ll_sb_info *sbi = ll_i2sbi(inode);
5641 struct lustre_handle lockh;
5642 struct layout_intent intent = {
5643 .li_opc = LAYOUT_INTENT_ACCESS,
5645 enum ldlm_mode mode;
5649 *gen = ll_layout_version_get(lli);
5650 if (!(sbi->ll_flags & LL_SBI_LAYOUT_LOCK) || *gen != CL_LAYOUT_GEN_NONE)
5654 LASSERT(fid_is_sane(ll_inode2fid(inode)));
5655 LASSERT(S_ISREG(inode->i_mode));
5657 /* take layout lock mutex to enqueue layout lock exclusively. */
5658 mutex_lock(&lli->lli_layout_mutex);
5661 /* mostly layout lock is caching on the local side, so try to
5662 * match it before grabbing layout lock mutex. */
5663 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
5664 LCK_CR | LCK_CW | LCK_PR |
5666 if (mode != 0) { /* hit cached lock */
5667 rc = ll_layout_lock_set(&lockh, mode, inode);
5673 rc = ll_layout_intent(inode, &intent);
5679 *gen = ll_layout_version_get(lli);
5680 mutex_unlock(&lli->lli_layout_mutex);
5686 * Issue layout intent RPC indicating where in a file an IO is about to write.
5688 * \param[in] inode file inode.
5689 * \param[in] ext write range with start offset of fille in bytes where
5690 * an IO is about to write, and exclusive end offset in
5693 * \retval 0 on success
5694 * \retval < 0 error code
5696 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
5697 struct lu_extent *ext)
5699 struct layout_intent intent = {
5701 .li_extent.e_start = ext->e_start,
5702 .li_extent.e_end = ext->e_end,
5707 rc = ll_layout_intent(inode, &intent);
5713 * This function send a restore request to the MDT
5715 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
5717 struct hsm_user_request *hur;
5721 len = sizeof(struct hsm_user_request) +
5722 sizeof(struct hsm_user_item);
5723 OBD_ALLOC(hur, len);
5727 hur->hur_request.hr_action = HUA_RESTORE;
5728 hur->hur_request.hr_archive_id = 0;
5729 hur->hur_request.hr_flags = 0;
5730 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
5731 sizeof(hur->hur_user_item[0].hui_fid));
5732 hur->hur_user_item[0].hui_extent.offset = offset;
5733 hur->hur_user_item[0].hui_extent.length = length;
5734 hur->hur_request.hr_itemcount = 1;
5735 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,