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/
33 * Author: Peter Braam <braam@clusterfs.com>
34 * Author: Phil Schwan <phil@clusterfs.com>
35 * Author: Andreas Dilger <adilger@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <lustre_dlm.h>
40 #include <linux/pagemap.h>
41 #include <linux/file.h>
42 #include <linux/sched.h>
43 #include <linux/user_namespace.h>
44 #include <linux/uidgid.h>
45 #include <linux/falloc.h>
46 #include <linux/ktime.h>
48 #include <uapi/linux/lustre/lustre_ioctl.h>
49 #include <lustre_swab.h>
51 #include "cl_object.h"
52 #include "llite_internal.h"
53 #include "vvp_internal.h"
56 struct inode *sp_inode;
61 __u64 pa_data_version;
67 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
69 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
72 static struct ll_file_data *ll_file_data_get(void)
74 struct ll_file_data *fd;
76 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
80 fd->fd_write_failed = false;
81 pcc_file_init(&fd->fd_pcc_file);
86 static void ll_file_data_put(struct ll_file_data *fd)
89 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
93 * Packs all the attributes into @op_data for the CLOSE rpc.
95 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
96 struct obd_client_handle *och)
100 ll_prep_md_op_data(op_data, inode, NULL, NULL,
101 0, 0, LUSTRE_OPC_ANY, NULL);
103 op_data->op_attr.ia_mode = inode->i_mode;
104 op_data->op_attr.ia_atime = inode->i_atime;
105 op_data->op_attr.ia_mtime = inode->i_mtime;
106 op_data->op_attr.ia_ctime = inode->i_ctime;
107 /* In case of encrypted file without the key, visible size was rounded
108 * up to next LUSTRE_ENCRYPTION_UNIT_SIZE, and clear text size was
109 * stored into lli_lazysize in ll_merge_attr(), so set proper file size
110 * now that we are closing.
112 if (llcrypt_require_key(inode) == -ENOKEY &&
113 ll_i2info(inode)->lli_attr_valid & OBD_MD_FLLAZYSIZE)
114 op_data->op_attr.ia_size = ll_i2info(inode)->lli_lazysize;
116 op_data->op_attr.ia_size = i_size_read(inode);
117 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
118 ATTR_MTIME | ATTR_MTIME_SET |
120 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
121 op_data->op_attr_blocks = inode->i_blocks;
122 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
123 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
124 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
125 op_data->op_open_handle = och->och_open_handle;
127 if (och->och_flags & FMODE_WRITE &&
128 test_and_clear_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags))
129 /* For HSM: if inode data has been modified, pack it so that
130 * MDT can set data dirty flag in the archive. */
131 op_data->op_bias |= MDS_DATA_MODIFIED;
137 * Perform a close, possibly with a bias.
138 * The meaning of "data" depends on the value of "bias".
140 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
141 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
144 static int ll_close_inode_openhandle(struct inode *inode,
145 struct obd_client_handle *och,
146 enum mds_op_bias bias, void *data)
148 struct obd_export *md_exp = ll_i2mdexp(inode);
149 const struct ll_inode_info *lli = ll_i2info(inode);
150 struct md_op_data *op_data;
151 struct ptlrpc_request *req = NULL;
155 if (class_exp2obd(md_exp) == NULL) {
156 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
157 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
161 OBD_ALLOC_PTR(op_data);
162 /* We leak openhandle and request here on error, but not much to be
163 * done in OOM case since app won't retry close on error either. */
165 GOTO(out, rc = -ENOMEM);
167 ll_prepare_close(inode, op_data, och);
169 case MDS_CLOSE_LAYOUT_MERGE:
170 /* merge blocks from the victim inode */
171 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
172 op_data->op_attr.ia_valid |= ATTR_SIZE;
173 op_data->op_xvalid |= OP_XVALID_BLOCKS;
175 case MDS_CLOSE_LAYOUT_SPLIT:
176 case MDS_CLOSE_LAYOUT_SWAP: {
177 struct split_param *sp = data;
179 LASSERT(data != NULL);
180 op_data->op_bias |= bias;
181 op_data->op_data_version = 0;
182 op_data->op_lease_handle = och->och_lease_handle;
183 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
184 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
185 op_data->op_mirror_id = sp->sp_mirror_id;
187 op_data->op_fid2 = *ll_inode2fid(data);
192 case MDS_CLOSE_RESYNC_DONE: {
193 struct ll_ioc_lease *ioc = data;
195 LASSERT(data != NULL);
196 op_data->op_attr_blocks +=
197 ioc->lil_count * op_data->op_attr_blocks;
198 op_data->op_attr.ia_valid |= ATTR_SIZE;
199 op_data->op_xvalid |= OP_XVALID_BLOCKS;
200 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
202 op_data->op_lease_handle = och->och_lease_handle;
203 op_data->op_data = &ioc->lil_ids[0];
204 op_data->op_data_size =
205 ioc->lil_count * sizeof(ioc->lil_ids[0]);
209 case MDS_PCC_ATTACH: {
210 struct pcc_param *param = data;
212 LASSERT(data != NULL);
213 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
214 op_data->op_archive_id = param->pa_archive_id;
215 op_data->op_data_version = param->pa_data_version;
216 op_data->op_lease_handle = och->och_lease_handle;
220 case MDS_HSM_RELEASE:
221 LASSERT(data != NULL);
222 op_data->op_bias |= MDS_HSM_RELEASE;
223 op_data->op_data_version = *(__u64 *)data;
224 op_data->op_lease_handle = och->och_lease_handle;
225 op_data->op_attr.ia_valid |= ATTR_SIZE;
226 op_data->op_xvalid |= OP_XVALID_BLOCKS;
230 LASSERT(data == NULL);
234 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
235 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
236 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
237 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
239 rc = md_close(md_exp, op_data, och->och_mod, &req);
240 if (rc != 0 && rc != -EINTR)
241 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
242 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
244 if (rc == 0 && op_data->op_bias & bias) {
245 struct mdt_body *body;
247 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
248 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
251 if (bias & MDS_PCC_ATTACH) {
252 struct pcc_param *param = data;
254 param->pa_layout_gen = body->mbo_layout_gen;
258 ll_finish_md_op_data(op_data);
262 md_clear_open_replay_data(md_exp, och);
263 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
266 ptlrpc_req_finished(req); /* This is close request */
270 int ll_md_real_close(struct inode *inode, fmode_t fmode)
272 struct ll_inode_info *lli = ll_i2info(inode);
273 struct obd_client_handle **och_p;
274 struct obd_client_handle *och;
279 if (fmode & FMODE_WRITE) {
280 och_p = &lli->lli_mds_write_och;
281 och_usecount = &lli->lli_open_fd_write_count;
282 } else if (fmode & FMODE_EXEC) {
283 och_p = &lli->lli_mds_exec_och;
284 och_usecount = &lli->lli_open_fd_exec_count;
286 LASSERT(fmode & FMODE_READ);
287 och_p = &lli->lli_mds_read_och;
288 och_usecount = &lli->lli_open_fd_read_count;
291 mutex_lock(&lli->lli_och_mutex);
292 if (*och_usecount > 0) {
293 /* There are still users of this handle, so skip
295 mutex_unlock(&lli->lli_och_mutex);
301 mutex_unlock(&lli->lli_och_mutex);
304 /* There might be a race and this handle may already
306 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
312 static int ll_md_close(struct inode *inode, struct file *file)
314 union ldlm_policy_data policy = {
315 .l_inodebits = { MDS_INODELOCK_OPEN },
317 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
318 struct ll_file_data *fd = file->private_data;
319 struct ll_inode_info *lli = ll_i2info(inode);
320 struct lustre_handle lockh;
321 enum ldlm_mode lockmode;
325 /* clear group lock, if present */
326 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
327 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
329 mutex_lock(&lli->lli_och_mutex);
330 if (fd->fd_lease_och != NULL) {
332 struct obd_client_handle *lease_och;
334 lease_och = fd->fd_lease_och;
335 fd->fd_lease_och = NULL;
336 mutex_unlock(&lli->lli_och_mutex);
338 /* Usually the lease is not released when the
339 * application crashed, we need to release here. */
340 rc = ll_lease_close(lease_och, inode, &lease_broken);
342 mutex_lock(&lli->lli_och_mutex);
344 CDEBUG_LIMIT(rc ? D_ERROR : D_INODE,
345 "Clean up lease "DFID" %d/%d\n",
346 PFID(&lli->lli_fid), rc, lease_broken);
349 if (fd->fd_och != NULL) {
350 struct obd_client_handle *och;
354 mutex_unlock(&lli->lli_och_mutex);
356 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
360 /* Let's see if we have good enough OPEN lock on the file and if
361 we can skip talking to MDS */
362 if (fd->fd_omode & FMODE_WRITE) {
364 LASSERT(lli->lli_open_fd_write_count);
365 lli->lli_open_fd_write_count--;
366 } else if (fd->fd_omode & FMODE_EXEC) {
368 LASSERT(lli->lli_open_fd_exec_count);
369 lli->lli_open_fd_exec_count--;
372 LASSERT(lli->lli_open_fd_read_count);
373 lli->lli_open_fd_read_count--;
375 mutex_unlock(&lli->lli_och_mutex);
377 /* LU-4398: do not cache write open lock if the file has exec bit */
378 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
379 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
380 LDLM_IBITS, &policy, lockmode, &lockh))
381 rc = ll_md_real_close(inode, fd->fd_omode);
384 file->private_data = NULL;
385 ll_file_data_put(fd);
390 /* While this returns an error code, fput() the caller does not, so we need
391 * to make every effort to clean up all of our state here. Also, applications
392 * rarely check close errors and even if an error is returned they will not
393 * re-try the close call.
395 int ll_file_release(struct inode *inode, struct file *file)
397 struct ll_file_data *fd;
398 struct ll_sb_info *sbi = ll_i2sbi(inode);
399 struct ll_inode_info *lli = ll_i2info(inode);
400 ktime_t kstart = ktime_get();
405 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
406 PFID(ll_inode2fid(inode)), inode);
408 fd = file->private_data;
411 /* The last ref on @file, maybe not the the owner pid of statahead,
412 * because parent and child process can share the same file handle. */
413 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
414 ll_deauthorize_statahead(inode, fd);
416 if (is_root_inode(inode)) {
417 file->private_data = NULL;
418 ll_file_data_put(fd);
422 pcc_file_release(inode, file);
424 if (!S_ISDIR(inode->i_mode)) {
425 if (lli->lli_clob != NULL)
426 lov_read_and_clear_async_rc(lli->lli_clob);
427 lli->lli_async_rc = 0;
430 lli->lli_close_fd_time = ktime_get();
432 rc = ll_md_close(inode, file);
434 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
435 libcfs_debug_dumplog();
438 if (!rc && !is_root_inode(inode))
439 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
440 ktime_us_delta(ktime_get(), kstart));
444 static inline int ll_dom_readpage(void *data, struct page *page)
446 /* since ll_dom_readpage is a page cache helper, it is safe to assume
447 * mapping and host pointers are set here
450 struct niobuf_local *lnb = data;
454 inode = page2inode(page);
456 kaddr = kmap_atomic(page);
457 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
458 if (lnb->lnb_len < PAGE_SIZE)
459 memset(kaddr + lnb->lnb_len, 0,
460 PAGE_SIZE - lnb->lnb_len);
461 flush_dcache_page(page);
462 SetPageUptodate(page);
463 kunmap_atomic(kaddr);
465 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
466 if (!llcrypt_has_encryption_key(inode))
467 CDEBUG(D_SEC, "no enc key for "DFID"\n",
468 PFID(ll_inode2fid(inode)));
470 unsigned int offs = 0;
472 while (offs < PAGE_SIZE) {
473 /* decrypt only if page is not empty */
474 if (memcmp(page_address(page) + offs,
475 page_address(ZERO_PAGE(0)),
476 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
479 rc = llcrypt_decrypt_pagecache_blocks(page,
480 LUSTRE_ENCRYPTION_UNIT_SIZE,
485 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
494 #ifdef HAVE_READ_CACHE_PAGE_WANTS_FILE
495 static inline int ll_dom_read_folio(struct file *file, struct folio *folio0)
497 return ll_dom_readpage(file->private_data, folio_page(folio0, 0));
500 #define ll_dom_read_folio ll_dom_readpage
503 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
507 struct ll_inode_info *lli = ll_i2info(inode);
508 struct cl_object *obj = lli->lli_clob;
509 struct address_space *mapping = inode->i_mapping;
511 struct niobuf_remote *rnb;
512 struct mdt_body *body;
514 unsigned long index, start;
515 struct niobuf_local lnb;
524 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
528 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
529 if (rnb == NULL || rnb->rnb_len == 0)
532 /* LU-11595: Server may return whole file and that is OK always or
533 * it may return just file tail and its offset must be aligned with
534 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
535 * smaller then offset may be not aligned and that data is just ignored.
537 if (rnb->rnb_offset & ~PAGE_MASK)
540 /* Server returns whole file or just file tail if it fills in reply
541 * buffer, in both cases total size should be equal to the file size.
543 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
544 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
545 !(inode && IS_ENCRYPTED(inode))) {
546 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
547 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
548 rnb->rnb_len, body->mbo_dom_size);
552 env = cl_env_get(&refcheck);
555 io = vvp_env_thread_io(env);
557 io->ci_ignore_layout = 1;
558 rc = cl_io_init(env, io, CIT_MISC, obj);
562 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
563 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
565 data = (char *)rnb + sizeof(*rnb);
567 lnb.lnb_file_offset = rnb->rnb_offset;
568 start = lnb.lnb_file_offset >> PAGE_SHIFT;
570 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
571 lnb.lnb_page_offset = 0;
573 struct cl_page *page;
575 lnb.lnb_data = data + (index << PAGE_SHIFT);
576 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
577 if (lnb.lnb_len > PAGE_SIZE)
578 lnb.lnb_len = PAGE_SIZE;
580 vmpage = ll_read_cache_page(mapping, index + start,
581 ll_dom_read_folio, &lnb);
582 if (IS_ERR(vmpage)) {
583 CWARN("%s: cannot fill page %lu for "DFID
584 " with data: rc = %li\n",
585 ll_i2sbi(inode)->ll_fsname, index + start,
586 PFID(lu_object_fid(&obj->co_lu)),
591 if (vmpage->mapping == NULL) {
594 /* page was truncated */
597 /* attach VM page to CL page cache */
598 page = cl_page_find(env, obj, vmpage->index, vmpage,
601 ClearPageUptodate(vmpage);
606 SetPageUptodate(vmpage);
607 cl_page_put(env, page);
611 } while (rnb->rnb_len > (index << PAGE_SHIFT));
615 cl_env_put(env, &refcheck);
620 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
621 struct lookup_intent *itp)
623 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
624 struct dentry *parent = de->d_parent;
627 struct md_op_data *op_data;
628 struct ptlrpc_request *req = NULL;
632 LASSERT(parent != NULL);
633 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
635 /* if server supports open-by-fid, or file name is invalid, don't pack
636 * name in open request */
637 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
638 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
640 len = de->d_name.len;
641 name = kmalloc(len + 1, GFP_NOFS);
646 spin_lock(&de->d_lock);
647 if (len != de->d_name.len) {
648 spin_unlock(&de->d_lock);
652 memcpy(name, de->d_name.name, len);
654 spin_unlock(&de->d_lock);
656 if (!lu_name_is_valid_2(name, len)) {
662 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
663 name, len, 0, LUSTRE_OPC_OPEN, NULL);
664 if (IS_ERR(op_data)) {
666 RETURN(PTR_ERR(op_data));
668 op_data->op_data = lmm;
669 op_data->op_data_size = lmmsize;
671 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_OPEN_DELAY, cfs_fail_val);
673 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
674 &ll_md_blocking_ast, 0);
676 ll_finish_md_op_data(op_data);
678 /* reason for keep own exit path - don`t flood log
679 * with messages with -ESTALE errors.
681 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
682 it_open_error(DISP_OPEN_OPEN, itp))
684 ll_release_openhandle(de, itp);
688 if (it_disposition(itp, DISP_LOOKUP_NEG))
689 GOTO(out, rc = -ENOENT);
691 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
692 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
693 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
697 rc = ll_prep_inode(&de->d_inode, &req->rq_pill, NULL, itp);
699 if (!rc && itp->it_lock_mode) {
702 /* If we got a lock back and it has a LOOKUP bit set,
703 * make sure the dentry is marked as valid so we can find it.
704 * We don't need to care about actual hashing since other bits
705 * of kernel will deal with that later.
707 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
708 if (bits & MDS_INODELOCK_LOOKUP) {
709 d_lustre_revalidate(de);
710 ll_update_dir_depth(parent->d_inode, de->d_inode);
713 /* if DoM bit returned along with LAYOUT bit then there
714 * can be read-on-open data returned.
716 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
717 ll_dom_finish_open(de->d_inode, req);
721 ptlrpc_req_finished(req);
722 ll_intent_drop_lock(itp);
724 /* We did open by fid, but by the time we got to the server, the object
725 * disappeared. This is possible if the object was unlinked, but it's
726 * also possible if the object was unlinked by a rename. In the case
727 * of an object renamed over our existing one, we can't fail this open.
728 * O_CREAT also goes through this path if we had an existing dentry,
729 * and it's obviously wrong to return ENOENT for O_CREAT.
731 * Instead let's return -ESTALE, and the VFS will retry the open with
732 * LOOKUP_REVAL, which we catch in ll_revalidate_dentry and fail to
733 * revalidate, causing a lookup. This causes extra lookups in the case
734 * where we had a dentry in cache but the file is being unlinked and we
735 * lose the race with unlink, but this should be very rare.
743 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
744 struct obd_client_handle *och)
746 struct mdt_body *body;
748 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
749 och->och_open_handle = body->mbo_open_handle;
750 och->och_fid = body->mbo_fid1;
751 och->och_lease_handle.cookie = it->it_lock_handle;
752 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
753 och->och_flags = it->it_flags;
755 return md_set_open_replay_data(md_exp, och, it);
758 static int ll_local_open(struct file *file, struct lookup_intent *it,
759 struct ll_file_data *fd, struct obd_client_handle *och)
761 struct inode *inode = file_inode(file);
764 LASSERT(!file->private_data);
771 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
776 file->private_data = fd;
777 ll_readahead_init(inode, &fd->fd_ras);
778 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
783 void ll_track_file_opens(struct inode *inode)
785 struct ll_inode_info *lli = ll_i2info(inode);
786 struct ll_sb_info *sbi = ll_i2sbi(inode);
788 /* do not skew results with delays from never-opened inodes */
789 if (ktime_to_ns(lli->lli_close_fd_time))
790 ll_stats_ops_tally(sbi, LPROC_LL_INODE_OPCLTM,
791 ktime_us_delta(ktime_get(), lli->lli_close_fd_time));
793 if (ktime_after(ktime_get(),
794 ktime_add_ms(lli->lli_close_fd_time,
795 sbi->ll_oc_max_ms))) {
796 lli->lli_open_fd_count = 1;
797 lli->lli_close_fd_time = ns_to_ktime(0);
799 lli->lli_open_fd_count++;
802 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_OCOUNT,
803 lli->lli_open_fd_count);
806 /* Open a file, and (for the very first open) create objects on the OSTs at
807 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
808 * creation or open until ll_lov_setstripe() ioctl is called.
810 * If we already have the stripe MD locally then we don't request it in
811 * md_open(), by passing a lmm_size = 0.
813 * It is up to the application to ensure no other processes open this file
814 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
815 * used. We might be able to avoid races of that sort by getting lli_open_sem
816 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
817 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
819 int ll_file_open(struct inode *inode, struct file *file)
821 struct ll_inode_info *lli = ll_i2info(inode);
822 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
823 .it_flags = file->f_flags };
824 struct obd_client_handle **och_p = NULL;
825 __u64 *och_usecount = NULL;
826 struct ll_file_data *fd;
827 ktime_t kstart = ktime_get();
831 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
832 PFID(ll_inode2fid(inode)), inode, file->f_flags);
834 it = file->private_data; /* XXX: compat macro */
835 file->private_data = NULL; /* prevent ll_local_open assertion */
837 if (S_ISREG(inode->i_mode)) {
838 rc = ll_file_open_encrypt(inode, file);
840 if (it && it->it_disposition)
841 ll_release_openhandle(file_dentry(file), it);
842 GOTO(out_nofiledata, rc);
846 fd = ll_file_data_get();
848 GOTO(out_nofiledata, rc = -ENOMEM);
851 if (S_ISDIR(inode->i_mode))
852 ll_authorize_statahead(inode, fd);
854 ll_track_file_opens(inode);
855 if (is_root_inode(inode)) {
856 file->private_data = fd;
860 if (!it || !it->it_disposition) {
861 /* Convert f_flags into access mode. We cannot use file->f_mode,
862 * because everything but O_ACCMODE mask was stripped from
864 if ((oit.it_flags + 1) & O_ACCMODE)
866 if (file->f_flags & O_TRUNC)
867 oit.it_flags |= FMODE_WRITE;
869 /* kernel only call f_op->open in dentry_open. filp_open calls
870 * dentry_open after call to open_namei that checks permissions.
871 * Only nfsd_open call dentry_open directly without checking
872 * permissions and because of that this code below is safe.
874 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
875 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
877 /* We do not want O_EXCL here, presumably we opened the file
878 * already? XXX - NFS implications? */
879 oit.it_flags &= ~O_EXCL;
881 /* bug20584, if "it_flags" contains O_CREAT, the file will be
882 * created if necessary, then "IT_CREAT" should be set to keep
883 * consistent with it */
884 if (oit.it_flags & O_CREAT)
885 oit.it_op |= IT_CREAT;
891 /* Let's see if we have file open on MDS already. */
892 if (it->it_flags & FMODE_WRITE) {
893 och_p = &lli->lli_mds_write_och;
894 och_usecount = &lli->lli_open_fd_write_count;
895 } else if (it->it_flags & FMODE_EXEC) {
896 och_p = &lli->lli_mds_exec_och;
897 och_usecount = &lli->lli_open_fd_exec_count;
899 och_p = &lli->lli_mds_read_och;
900 och_usecount = &lli->lli_open_fd_read_count;
903 mutex_lock(&lli->lli_och_mutex);
904 if (*och_p) { /* Open handle is present */
905 if (it_disposition(it, DISP_OPEN_OPEN)) {
906 /* Well, there's extra open request that we do not need,
907 * let's close it somehow. This will decref request. */
908 rc = it_open_error(DISP_OPEN_OPEN, it);
910 mutex_unlock(&lli->lli_och_mutex);
911 GOTO(out_openerr, rc);
914 ll_release_openhandle(file_dentry(file), it);
918 rc = ll_local_open(file, it, fd, NULL);
921 mutex_unlock(&lli->lli_och_mutex);
922 GOTO(out_openerr, rc);
925 LASSERT(*och_usecount == 0);
926 if (!it->it_disposition) {
927 struct dentry *dentry = file_dentry(file);
928 struct ll_sb_info *sbi = ll_i2sbi(inode);
929 int open_threshold = sbi->ll_oc_thrsh_count;
931 /* We cannot just request lock handle now, new ELC code
932 * means that one of other OPEN locks for this file
933 * could be cancelled, and since blocking ast handler
934 * would attempt to grab och_mutex as well, that would
935 * result in a deadlock
937 mutex_unlock(&lli->lli_och_mutex);
939 * Normally called under two situations:
940 * 1. fhandle / NFS export.
941 * 2. A race/condition on MDS resulting in no open
942 * handle to be returned from LOOKUP|OPEN request,
943 * for example if the target entry was a symlink.
945 * For NFSv3 we need to always cache the open lock
946 * for pre 5.5 Linux kernels.
948 * After reaching number of opens of this inode
949 * we always ask for an open lock on it to handle
950 * bad userspace actors that open and close files
951 * in a loop for absolutely no good reason
953 /* fhandle / NFS path. */
954 if (lli->lli_open_thrsh_count != UINT_MAX)
955 open_threshold = lli->lli_open_thrsh_count;
957 if (filename_is_volatile(dentry->d_name.name,
960 /* There really is nothing here, but this
961 * make this more readable I think.
962 * We do not want openlock for volatile
963 * files under any circumstances
965 } else if (open_threshold > 0) {
966 /* Take MDS_OPEN_LOCK with many opens */
967 if (lli->lli_open_fd_count >= open_threshold)
968 it->it_flags |= MDS_OPEN_LOCK;
970 /* If this is open after we just closed */
971 else if (ktime_before(ktime_get(),
972 ktime_add_ms(lli->lli_close_fd_time,
973 sbi->ll_oc_thrsh_ms)))
974 it->it_flags |= MDS_OPEN_LOCK;
978 * Always specify MDS_OPEN_BY_FID because we don't want
979 * to get file with different fid.
981 it->it_flags |= MDS_OPEN_BY_FID;
982 rc = ll_intent_file_open(dentry, NULL, 0, it);
984 GOTO(out_openerr, rc);
988 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
990 GOTO(out_och_free, rc = -ENOMEM);
994 /* md_intent_lock() didn't get a request ref if there was an
995 * open error, so don't do cleanup on the request here
997 /* XXX (green): Should not we bail out on any error here, not
998 * just open error? */
999 rc = it_open_error(DISP_OPEN_OPEN, it);
1001 GOTO(out_och_free, rc);
1003 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
1004 "inode %p: disposition %x, status %d\n", inode,
1005 it_disposition(it, ~0), it->it_status);
1007 rc = ll_local_open(file, it, fd, *och_p);
1009 GOTO(out_och_free, rc);
1012 rc = pcc_file_open(inode, file);
1014 GOTO(out_och_free, rc);
1016 mutex_unlock(&lli->lli_och_mutex);
1020 /* Must do this outside lli_och_mutex lock to prevent deadlock where
1021 different kind of OPEN lock for this same inode gets cancelled
1022 by ldlm_cancel_lru */
1023 if (!S_ISREG(inode->i_mode))
1024 GOTO(out_och_free, rc);
1025 cl_lov_delay_create_clear(&file->f_flags);
1026 GOTO(out_och_free, rc);
1030 if (och_p && *och_p) {
1031 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
1032 *och_p = NULL; /* OBD_FREE writes some magic there */
1035 mutex_unlock(&lli->lli_och_mutex);
1038 if (lli->lli_opendir_key == fd)
1039 ll_deauthorize_statahead(inode, fd);
1042 ll_file_data_put(fd);
1044 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
1045 ktime_us_delta(ktime_get(), kstart));
1049 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
1050 ptlrpc_req_finished(it->it_request);
1051 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1057 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
1058 struct ldlm_lock_desc *desc, void *data, int flag)
1061 struct lustre_handle lockh;
1065 case LDLM_CB_BLOCKING:
1066 ldlm_lock2handle(lock, &lockh);
1067 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
1069 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1073 case LDLM_CB_CANCELING:
1081 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1082 * and save it as fd->fd_och so as to force client to reopen the file even
1083 * if it has an open lock in cache already.
1085 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1086 struct lustre_handle *old_open_handle)
1088 struct ll_inode_info *lli = ll_i2info(inode);
1089 struct ll_file_data *fd = file->private_data;
1090 struct obd_client_handle **och_p;
1091 __u64 *och_usecount;
1095 /* Get the openhandle of the file */
1096 mutex_lock(&lli->lli_och_mutex);
1097 if (fd->fd_lease_och != NULL)
1098 GOTO(out_unlock, rc = -EBUSY);
1100 if (fd->fd_och == NULL) {
1101 if (file->f_mode & FMODE_WRITE) {
1102 LASSERT(lli->lli_mds_write_och != NULL);
1103 och_p = &lli->lli_mds_write_och;
1104 och_usecount = &lli->lli_open_fd_write_count;
1106 LASSERT(lli->lli_mds_read_och != NULL);
1107 och_p = &lli->lli_mds_read_och;
1108 och_usecount = &lli->lli_open_fd_read_count;
1111 if (*och_usecount > 1)
1112 GOTO(out_unlock, rc = -EBUSY);
1114 fd->fd_och = *och_p;
1119 *old_open_handle = fd->fd_och->och_open_handle;
1123 mutex_unlock(&lli->lli_och_mutex);
1128 * Release ownership on lli_mds_*_och when putting back a file lease.
1130 static int ll_lease_och_release(struct inode *inode, struct file *file)
1132 struct ll_inode_info *lli = ll_i2info(inode);
1133 struct ll_file_data *fd = file->private_data;
1134 struct obd_client_handle **och_p;
1135 struct obd_client_handle *old_och = NULL;
1136 __u64 *och_usecount;
1140 mutex_lock(&lli->lli_och_mutex);
1141 if (file->f_mode & FMODE_WRITE) {
1142 och_p = &lli->lli_mds_write_och;
1143 och_usecount = &lli->lli_open_fd_write_count;
1145 och_p = &lli->lli_mds_read_och;
1146 och_usecount = &lli->lli_open_fd_read_count;
1149 /* The file may have been open by another process (broken lease) so
1150 * *och_p is not NULL. In this case we should simply increase usecount
1153 if (*och_p != NULL) {
1154 old_och = fd->fd_och;
1157 *och_p = fd->fd_och;
1161 mutex_unlock(&lli->lli_och_mutex);
1163 if (old_och != NULL)
1164 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1170 * Acquire a lease and open the file.
1172 static struct obd_client_handle *
1173 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1176 struct lookup_intent it = { .it_op = IT_OPEN };
1177 struct ll_sb_info *sbi = ll_i2sbi(inode);
1178 struct md_op_data *op_data;
1179 struct ptlrpc_request *req = NULL;
1180 struct lustre_handle old_open_handle = { 0 };
1181 struct obd_client_handle *och = NULL;
1186 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1187 RETURN(ERR_PTR(-EINVAL));
1190 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1191 RETURN(ERR_PTR(-EPERM));
1193 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1195 RETURN(ERR_PTR(rc));
1200 RETURN(ERR_PTR(-ENOMEM));
1202 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1203 LUSTRE_OPC_ANY, NULL);
1204 if (IS_ERR(op_data))
1205 GOTO(out, rc = PTR_ERR(op_data));
1207 /* To tell the MDT this openhandle is from the same owner */
1208 op_data->op_open_handle = old_open_handle;
1210 it.it_flags = fmode | open_flags;
1211 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1212 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1213 &ll_md_blocking_lease_ast,
1214 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1215 * it can be cancelled which may mislead applications that the lease is
1217 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1218 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1219 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1220 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1221 ll_finish_md_op_data(op_data);
1222 ptlrpc_req_finished(req);
1224 GOTO(out_release_it, rc);
1226 if (it_disposition(&it, DISP_LOOKUP_NEG))
1227 GOTO(out_release_it, rc = -ENOENT);
1229 rc = it_open_error(DISP_OPEN_OPEN, &it);
1231 GOTO(out_release_it, rc);
1233 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1234 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1236 GOTO(out_release_it, rc);
1238 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1239 GOTO(out_close, rc = -EOPNOTSUPP);
1241 /* already get lease, handle lease lock */
1242 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1243 if (!it.it_lock_mode ||
1244 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1245 /* open lock must return for lease */
1246 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1247 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1249 GOTO(out_close, rc = -EPROTO);
1252 ll_intent_release(&it);
1256 /* Cancel open lock */
1257 if (it.it_lock_mode != 0) {
1258 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1260 it.it_lock_mode = 0;
1261 och->och_lease_handle.cookie = 0ULL;
1263 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1265 CERROR("%s: error closing file "DFID": %d\n",
1266 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1267 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1269 ll_intent_release(&it);
1273 RETURN(ERR_PTR(rc));
1277 * Check whether a layout swap can be done between two inodes.
1279 * \param[in] inode1 First inode to check
1280 * \param[in] inode2 Second inode to check
1282 * \retval 0 on success, layout swap can be performed between both inodes
1283 * \retval negative error code if requirements are not met
1285 static int ll_check_swap_layouts_validity(struct inode *inode1,
1286 struct inode *inode2)
1288 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1291 if (inode_permission(&init_user_ns, inode1, MAY_WRITE) ||
1292 inode_permission(&init_user_ns, inode2, MAY_WRITE))
1295 if (inode1->i_sb != inode2->i_sb)
1301 static int ll_swap_layouts_close(struct obd_client_handle *och,
1302 struct inode *inode, struct inode *inode2)
1304 const struct lu_fid *fid1 = ll_inode2fid(inode);
1305 const struct lu_fid *fid2;
1309 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1310 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1312 rc = ll_check_swap_layouts_validity(inode, inode2);
1314 GOTO(out_free_och, rc);
1316 /* We now know that inode2 is a lustre inode */
1317 fid2 = ll_inode2fid(inode2);
1319 rc = lu_fid_cmp(fid1, fid2);
1321 GOTO(out_free_och, rc = -EINVAL);
1323 /* Close the file and {swap,merge} layouts between inode & inode2.
1324 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1325 * because we still need it to pack l_remote_handle to MDT. */
1326 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1329 och = NULL; /* freed in ll_close_inode_openhandle() */
1339 * Release lease and close the file.
1340 * It will check if the lease has ever broken.
1342 static int ll_lease_close_intent(struct obd_client_handle *och,
1343 struct inode *inode,
1344 bool *lease_broken, enum mds_op_bias bias,
1347 struct ldlm_lock *lock;
1348 bool cancelled = true;
1352 lock = ldlm_handle2lock(&och->och_lease_handle);
1354 lock_res_and_lock(lock);
1355 cancelled = ldlm_is_cancel(lock);
1356 unlock_res_and_lock(lock);
1357 LDLM_LOCK_PUT(lock);
1360 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1361 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1363 if (lease_broken != NULL)
1364 *lease_broken = cancelled;
1366 if (!cancelled && !bias)
1367 ldlm_cli_cancel(&och->och_lease_handle, 0);
1369 if (cancelled) { /* no need to excute intent */
1374 rc = ll_close_inode_openhandle(inode, och, bias, data);
1378 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1381 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1385 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1387 static int ll_lease_file_resync(struct obd_client_handle *och,
1388 struct inode *inode, void __user *uarg)
1390 struct ll_sb_info *sbi = ll_i2sbi(inode);
1391 struct md_op_data *op_data;
1392 struct ll_ioc_lease_id ioc;
1393 __u64 data_version_unused;
1397 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1398 LUSTRE_OPC_ANY, NULL);
1399 if (IS_ERR(op_data))
1400 RETURN(PTR_ERR(op_data));
1402 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
1405 /* before starting file resync, it's necessary to clean up page cache
1406 * in client memory, otherwise once the layout version is increased,
1407 * writing back cached data will be denied the OSTs. */
1408 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1412 op_data->op_lease_handle = och->och_lease_handle;
1413 op_data->op_mirror_id = ioc.lil_mirror_id;
1414 rc = md_file_resync(sbi->ll_md_exp, op_data);
1420 ll_finish_md_op_data(op_data);
1424 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1426 struct ll_inode_info *lli = ll_i2info(inode);
1427 struct cl_object *obj = lli->lli_clob;
1428 struct cl_attr *attr = vvp_env_thread_attr(env);
1436 ll_inode_size_lock(inode);
1438 /* Merge timestamps the most recently obtained from MDS with
1439 * timestamps obtained from OSTs.
1441 * Do not overwrite atime of inode because it may be refreshed
1442 * by file_accessed() function. If the read was served by cache
1443 * data, there is no RPC to be sent so that atime may not be
1444 * transferred to OSTs at all. MDT only updates atime at close time
1445 * if it's at least 'mdd.*.atime_diff' older.
1446 * All in all, the atime in Lustre does not strictly comply with
1447 * POSIX. Solving this problem needs to send an RPC to MDT for each
1448 * read, this will hurt performance.
1450 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1451 inode->i_atime.tv_sec < lli->lli_atime)
1452 inode->i_atime.tv_sec = lli->lli_atime;
1454 inode->i_mtime.tv_sec = lli->lli_mtime;
1455 inode->i_ctime.tv_sec = lli->lli_ctime;
1457 mtime = inode->i_mtime.tv_sec;
1458 atime = inode->i_atime.tv_sec;
1459 ctime = inode->i_ctime.tv_sec;
1461 cl_object_attr_lock(obj);
1462 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1465 rc = cl_object_attr_get(env, obj, attr);
1466 cl_object_attr_unlock(obj);
1469 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1471 if (atime < attr->cat_atime)
1472 atime = attr->cat_atime;
1474 if (ctime < attr->cat_ctime)
1475 ctime = attr->cat_ctime;
1477 if (mtime < attr->cat_mtime)
1478 mtime = attr->cat_mtime;
1480 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1481 PFID(&lli->lli_fid), attr->cat_size);
1483 if (llcrypt_require_key(inode) == -ENOKEY) {
1484 /* Without the key, round up encrypted file size to next
1485 * LUSTRE_ENCRYPTION_UNIT_SIZE. Clear text size is put in
1486 * lli_lazysize for proper file size setting at close time.
1488 lli->lli_attr_valid |= OBD_MD_FLLAZYSIZE;
1489 lli->lli_lazysize = attr->cat_size;
1490 attr->cat_size = round_up(attr->cat_size,
1491 LUSTRE_ENCRYPTION_UNIT_SIZE);
1493 i_size_write(inode, attr->cat_size);
1494 inode->i_blocks = attr->cat_blocks;
1496 inode->i_mtime.tv_sec = mtime;
1497 inode->i_atime.tv_sec = atime;
1498 inode->i_ctime.tv_sec = ctime;
1501 ll_inode_size_unlock(inode);
1507 * Set designated mirror for I/O.
1509 * So far only read, write, and truncated can support to issue I/O to
1510 * designated mirror.
1512 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1514 struct ll_file_data *fd = file->private_data;
1516 /* clear layout version for generic(non-resync) I/O in case it carries
1517 * stale layout version due to I/O restart */
1518 io->ci_layout_version = 0;
1520 /* FLR: disable non-delay for designated mirror I/O because obviously
1521 * only one mirror is available */
1522 if (fd->fd_designated_mirror > 0) {
1524 io->ci_designated_mirror = fd->fd_designated_mirror;
1525 io->ci_layout_version = fd->fd_layout_version;
1528 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1529 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1533 * This is relatime_need_update() from Linux 5.17, which is not exported.
1535 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1536 struct timespec64 now)
1539 if (!(mnt->mnt_flags & MNT_RELATIME))
1542 * Is mtime younger than atime? If yes, update atime:
1544 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1547 * Is ctime younger than atime? If yes, update atime:
1549 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1553 * Is the previous atime value older than a day? If yes,
1556 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1559 * Good, we can skip the atime update:
1565 * Very similar to kernel function: !__atime_needs_update()
1567 static bool file_is_noatime(const struct file *file)
1569 struct vfsmount *mnt = file->f_path.mnt;
1570 struct inode *inode = file_inode((struct file *)file);
1571 struct timespec64 now;
1573 if (file->f_flags & O_NOATIME)
1576 if (inode->i_flags & S_NOATIME)
1579 if (IS_NOATIME(inode))
1582 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1585 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1588 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1591 now = current_time(inode);
1593 if (!relatime_need_update(mnt, inode, now))
1599 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1600 struct vvp_io_args *args)
1602 struct inode *inode = file_inode(file);
1603 struct ll_file_data *fd = file->private_data;
1605 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1606 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1608 if (iot == CIT_WRITE) {
1609 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1610 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1611 file->f_flags & O_DIRECT ||
1613 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1614 io->u.ci_wr.wr_sync |= !!(args &&
1615 (args->u.normal.via_iocb->ki_flags &
1621 io->ci_iocb_nowait = !!(args &&
1622 (args->u.normal.via_iocb->ki_flags &
1626 io->ci_obj = ll_i2info(inode)->lli_clob;
1627 io->ci_lockreq = CILR_MAYBE;
1628 if (ll_file_nolock(file)) {
1629 io->ci_lockreq = CILR_NEVER;
1630 io->ci_no_srvlock = 1;
1631 } else if (file->f_flags & O_APPEND) {
1632 io->ci_lockreq = CILR_MANDATORY;
1634 io->ci_noatime = file_is_noatime(file);
1635 io->ci_async_readahead = false;
1637 /* FLR: only use non-delay I/O for read as there is only one
1638 * avaliable mirror for write. */
1639 io->ci_ndelay = !(iot == CIT_WRITE);
1641 ll_io_set_mirror(io, file);
1644 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1647 struct ll_inode_info *lli = ll_i2info(inode);
1648 struct ll_sb_info *sbi = ll_i2sbi(inode);
1649 enum obd_heat_type sample_type;
1650 enum obd_heat_type iobyte_type;
1651 __u64 now = ktime_get_real_seconds();
1653 if (!ll_sbi_has_file_heat(sbi) ||
1654 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1657 if (iot == CIT_READ) {
1658 sample_type = OBD_HEAT_READSAMPLE;
1659 iobyte_type = OBD_HEAT_READBYTE;
1660 } else if (iot == CIT_WRITE) {
1661 sample_type = OBD_HEAT_WRITESAMPLE;
1662 iobyte_type = OBD_HEAT_WRITEBYTE;
1667 spin_lock(&lli->lli_heat_lock);
1668 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1669 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1670 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1671 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1672 spin_unlock(&lli->lli_heat_lock);
1676 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1677 struct file *file, enum cl_io_type iot,
1678 loff_t *ppos, size_t count)
1680 struct vvp_io *vio = vvp_env_io(env);
1681 struct inode *inode = file_inode(file);
1682 struct ll_inode_info *lli = ll_i2info(inode);
1683 struct ll_sb_info *sbi = ll_i2sbi(inode);
1684 struct ll_file_data *fd = file->private_data;
1685 struct range_lock range;
1686 bool range_locked = false;
1692 unsigned int retried = 0, dio_lock = 0;
1693 bool is_aio = false;
1694 bool is_parallel_dio = false;
1695 struct cl_dio_aio *ci_dio_aio = NULL;
1697 bool partial_io = false;
1698 size_t max_io_pages, max_cached_pages;
1702 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1703 file_dentry(file)->d_name.name,
1704 iot == CIT_READ ? "read" : "write", *ppos, count);
1706 max_io_pages = PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT;
1707 max_cached_pages = sbi->ll_cache->ccc_lru_max;
1708 if (max_io_pages > (max_cached_pages >> 2))
1709 max_io_pages = max_cached_pages >> 2;
1711 io = vvp_env_thread_io(env);
1712 if (file->f_flags & O_DIRECT) {
1713 if (file->f_flags & O_APPEND)
1715 if (!is_sync_kiocb(args->u.normal.via_iocb))
1718 /* the kernel does not support AIO on pipes, and parallel DIO
1719 * uses part of the AIO path, so we must not do parallel dio
1722 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1725 if (!ll_sbi_has_parallel_dio(sbi))
1726 is_parallel_dio = false;
1728 ci_dio_aio = cl_dio_aio_alloc(args->u.normal.via_iocb,
1729 ll_i2info(inode)->lli_clob, is_aio);
1731 GOTO(out, rc = -ENOMEM);
1736 * IO block size need be aware of cached page limit, otherwise
1737 * if we have small max_cached_mb but large block IO issued, io
1738 * could not be finished and blocked whole client.
1740 if (file->f_flags & O_DIRECT)
1743 per_bytes = min(max_io_pages << PAGE_SHIFT, count);
1744 partial_io = per_bytes < count;
1745 io = vvp_env_thread_io(env);
1746 ll_io_init(io, file, iot, args);
1747 io->ci_dio_aio = ci_dio_aio;
1748 io->ci_dio_lock = dio_lock;
1749 io->ci_ndelay_tried = retried;
1750 io->ci_parallel_dio = is_parallel_dio;
1752 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1753 if (file->f_flags & O_APPEND)
1754 range_lock_init(&range, 0, LUSTRE_EOF);
1756 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1758 vio->vui_fd = file->private_data;
1759 vio->vui_iter = args->u.normal.via_iter;
1760 vio->vui_iocb = args->u.normal.via_iocb;
1761 /* Direct IO reads must also take range lock,
1762 * or multiple reads will try to work on the same pages
1763 * See LU-6227 for details.
1765 if (((iot == CIT_WRITE) ||
1766 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1767 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1768 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1770 rc = range_lock(&lli->lli_write_tree, &range);
1774 range_locked = true;
1777 ll_cl_add(inode, env, io, LCC_RW);
1778 rc = cl_io_loop(env, io);
1779 ll_cl_remove(inode, env);
1781 /* cl_io_rw_init() handled IO */
1785 if (io->ci_dio_aio && !is_aio) {
1786 struct cl_sync_io *anchor = &io->ci_dio_aio->cda_sync;
1788 /* for dio, EIOCBQUEUED is an implementation detail,
1789 * and we don't return it to userspace
1791 if (rc == -EIOCBQUEUED)
1794 /* N/B: parallel DIO may be disabled during i/o submission;
1795 * if that occurs, I/O shifts to sync, so it's all resolved
1796 * before we get here, and this wait call completes
1799 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1805 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1807 range_unlock(&lli->lli_write_tree, &range);
1808 range_locked = false;
1812 * In order to move forward AIO, ci_nob was increased,
1813 * but that doesn't mean io have been finished, it just
1814 * means io have been submited, we will always return
1815 * EIOCBQUEUED to the caller, So we could only return
1816 * number of bytes in non-AIO case.
1818 if (io->ci_nob > 0) {
1821 result += io->ci_nob;
1822 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1827 count -= io->ci_nob;
1829 /* prepare IO restart */
1831 args->u.normal.via_iter = vio->vui_iter;
1835 * Reexpand iov count because it was zero
1838 iov_iter_reexpand(vio->vui_iter, count);
1839 if (per_bytes == io->ci_nob)
1840 io->ci_need_restart = 1;
1844 cl_io_fini(env, io);
1847 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1848 file->f_path.dentry->d_name.name,
1849 iot, rc, result, io->ci_need_restart);
1851 if ((!rc || rc == -ENODATA || rc == -ENOLCK || rc == -EIOCBQUEUED) &&
1852 count > 0 && io->ci_need_restart && retries-- > 0) {
1854 "%s: restart %s from ppos=%lld count=%zu retries=%u ret=%zd: rc = %d\n",
1855 file_dentry(file)->d_name.name,
1856 iot == CIT_READ ? "read" : "write",
1857 *ppos, count, retries, result, rc);
1858 /* preserve the tried count for FLR */
1859 retried = io->ci_ndelay_tried;
1860 dio_lock = io->ci_dio_lock;
1864 if (io->ci_dio_aio) {
1866 * VFS will call aio_complete() if no -EIOCBQUEUED
1867 * is returned for AIO, so we can not call aio_complete()
1870 * NB: This is safe because the atomic_dec_and_lock in
1871 * cl_sync_io_init has implicit memory barriers, so this will
1872 * be seen by whichever thread completes the DIO/AIO, even if
1875 if (rc != -EIOCBQUEUED)
1876 io->ci_dio_aio->cda_no_aio_complete = 1;
1878 * Drop one extra reference so that end_io() could be
1879 * called for this IO context, we could call it after
1880 * we make sure all AIO requests have been proceed.
1882 cl_sync_io_note(env, &io->ci_dio_aio->cda_sync,
1883 rc == -EIOCBQUEUED ? 0 : rc);
1885 LASSERT(io->ci_dio_aio->cda_creator_free);
1886 cl_dio_aio_free(env, io->ci_dio_aio);
1887 io->ci_dio_aio = NULL;
1891 if (iot == CIT_READ) {
1893 ll_stats_ops_tally(ll_i2sbi(inode),
1894 LPROC_LL_READ_BYTES, result);
1895 } else if (iot == CIT_WRITE) {
1897 ll_stats_ops_tally(ll_i2sbi(inode),
1898 LPROC_LL_WRITE_BYTES, result);
1899 fd->fd_write_failed = false;
1900 } else if (result == 0 && rc == 0) {
1903 fd->fd_write_failed = true;
1905 fd->fd_write_failed = false;
1906 } else if (rc != -ERESTARTSYS) {
1907 fd->fd_write_failed = true;
1911 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1913 ll_heat_add(inode, iot, result);
1915 RETURN(result > 0 ? result : rc);
1919 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1920 * especially for small I/O.
1922 * To serve a read request, CLIO has to create and initialize a cl_io and
1923 * then request DLM lock. This has turned out to have siginificant overhead
1924 * and affects the performance of small I/O dramatically.
1926 * It's not necessary to create a cl_io for each I/O. Under the help of read
1927 * ahead, most of the pages being read are already in memory cache and we can
1928 * read those pages directly because if the pages exist, the corresponding DLM
1929 * lock must exist so that page content must be valid.
1931 * In fast read implementation, the llite speculatively finds and reads pages
1932 * in memory cache. There are three scenarios for fast read:
1933 * - If the page exists and is uptodate, kernel VM will provide the data and
1934 * CLIO won't be intervened;
1935 * - If the page was brought into memory by read ahead, it will be exported
1936 * and read ahead parameters will be updated;
1937 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1938 * it will go back and invoke normal read, i.e., a cl_io will be created
1939 * and DLM lock will be requested.
1941 * POSIX compliance: posix standard states that read is intended to be atomic.
1942 * Lustre read implementation is in line with Linux kernel read implementation
1943 * and neither of them complies with POSIX standard in this matter. Fast read
1944 * doesn't make the situation worse on single node but it may interleave write
1945 * results from multiple nodes due to short read handling in ll_file_aio_read().
1947 * \param env - lu_env
1948 * \param iocb - kiocb from kernel
1949 * \param iter - user space buffers where the data will be copied
1951 * \retval - number of bytes have been read, or error code if error occurred.
1954 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1958 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1961 /* NB: we can't do direct IO for fast read because it will need a lock
1962 * to make IO engine happy. */
1963 if (iocb->ki_filp->f_flags & O_DIRECT)
1966 result = generic_file_read_iter(iocb, iter);
1968 /* If the first page is not in cache, generic_file_aio_read() will be
1969 * returned with -ENODATA.
1970 * See corresponding code in ll_readpage(). */
1971 if (result == -ENODATA)
1975 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1976 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1977 LPROC_LL_READ_BYTES, result);
1984 * Confine read iter lest read beyond the EOF
1986 * \param iocb [in] kernel iocb
1987 * \param to [in] reader iov_iter
1989 * \retval <0 failure
1991 * \retval >0 @iocb->ki_pos has passed the EOF
1993 static int file_read_confine_iter(struct lu_env *env, struct kiocb *iocb,
1994 struct iov_iter *to)
1996 struct cl_attr *attr = vvp_env_thread_attr(env);
1997 struct file *file = iocb->ki_filp;
1998 struct inode *inode = file_inode(file);
1999 struct ll_inode_info *lli = ll_i2info(inode);
2000 loff_t read_end = iocb->ki_pos + iov_iter_count(to);
2005 cl_object_attr_lock(lli->lli_clob);
2006 rc = cl_object_attr_get(env, lli->lli_clob, attr);
2007 cl_object_attr_unlock(lli->lli_clob);
2011 kms = attr->cat_kms;
2012 /* if read beyond end-of-file, adjust read count */
2013 if (kms > 0 && (iocb->ki_pos >= kms || read_end > kms)) {
2014 rc = ll_glimpse_size(inode);
2018 size = i_size_read(inode);
2019 if (iocb->ki_pos >= size || read_end > size) {
2021 "%s: read [%llu, %llu] over eof, kms %llu, file_size %llu.\n",
2022 file_dentry(file)->d_name.name,
2023 iocb->ki_pos, read_end, kms, size);
2025 if (iocb->ki_pos >= size)
2028 if (read_end > size)
2029 iov_iter_truncate(to, size - iocb->ki_pos);
2037 * Read from a file (through the page cache).
2039 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2042 struct vvp_io_args *args;
2043 struct file *file = iocb->ki_filp;
2047 ktime_t kstart = ktime_get();
2049 bool stale_data = false;
2053 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2054 file_dentry(file)->d_name.name,
2055 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2056 iov_iter_count(to));
2058 if (!iov_iter_count(to))
2061 env = cl_env_get(&refcheck);
2063 RETURN(PTR_ERR(env));
2065 result = file_read_confine_iter(env, iocb, to);
2068 else if (result > 0)
2072 * Currently when PCC read failed, we do not fall back to the
2073 * normal read path, just return the error.
2074 * The resaon is that: for RW-PCC, the file data may be modified
2075 * in the PCC and inconsistent with the data on OSTs (or file
2076 * data has been removed from the Lustre file system), at this
2077 * time, fallback to the normal read path may read the wrong
2079 * TODO: for RO-PCC (readonly PCC), fall back to normal read
2080 * path: read data from data copy on OSTs.
2082 result = pcc_file_read_iter(iocb, to, &cached);
2086 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
2088 result = ll_do_fast_read(iocb, to);
2089 if (result < 0 || iov_iter_count(to) == 0)
2092 args = ll_env_args(env);
2093 args->u.normal.via_iter = to;
2094 args->u.normal.via_iocb = iocb;
2096 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
2097 &iocb->ki_pos, iov_iter_count(to));
2100 else if (result == 0)
2104 cl_env_put(env, &refcheck);
2106 if (stale_data && result > 0) {
2108 * we've reached EOF before the read, the data read are cached
2111 iov_iter_truncate(to, 0);
2116 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2117 file->private_data, iocb->ki_pos, result,
2119 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
2120 ktime_us_delta(ktime_get(), kstart));
2124 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2125 file_dentry(file)->d_name.name,
2126 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2127 iov_iter_count(to));
2133 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
2134 * If a page is already in the page cache and dirty (and some other things -
2135 * See ll_tiny_write_begin for the instantiation of these rules), then we can
2136 * write to it without doing a full I/O, because Lustre already knows about it
2137 * and will write it out. This saves a lot of processing time.
2139 * All writes here are within one page, so exclusion is handled by the page
2140 * lock on the vm page. We do not do tiny writes for writes which touch
2141 * multiple pages because it's very unlikely multiple sequential pages are
2142 * are already dirty.
2144 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
2145 * and are unlikely to be to already dirty pages.
2147 * Attribute updates are important here, we do them in ll_tiny_write_end.
2149 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
2151 ssize_t count = iov_iter_count(iter);
2152 struct file *file = iocb->ki_filp;
2153 struct inode *inode = file_inode(file);
2154 bool lock_inode = !IS_NOSEC(inode);
2159 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2160 * of function for why.
2162 if (count >= PAGE_SIZE ||
2163 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2166 if (unlikely(lock_inode))
2168 result = __generic_file_write_iter(iocb, iter);
2170 if (unlikely(lock_inode))
2171 inode_unlock(inode);
2173 /* If the page is not already dirty, ll_tiny_write_begin returns
2174 * -ENODATA. We continue on to normal write.
2176 if (result == -ENODATA)
2180 ll_heat_add(inode, CIT_WRITE, result);
2181 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2184 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2190 * Write to a file (through the page cache).
2192 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2194 struct vvp_io_args *args;
2196 ssize_t rc_tiny = 0, rc_normal;
2197 struct file *file = iocb->ki_filp;
2200 ktime_t kstart = ktime_get();
2205 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2206 file_dentry(file)->d_name.name,
2207 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2208 iov_iter_count(from));
2210 if (!iov_iter_count(from))
2211 GOTO(out, rc_normal = 0);
2214 * When PCC write failed, we usually do not fall back to the normal
2215 * write path, just return the error. But there is a special case when
2216 * returned error code is -ENOSPC due to running out of space on PCC HSM
2217 * bakcend. At this time, it will fall back to normal I/O path and
2218 * retry the I/O. As the file is in HSM released state, it will restore
2219 * the file data to OSTs first and redo the write again. And the
2220 * restore process will revoke the layout lock and detach the file
2221 * from PCC cache automatically.
2223 result = pcc_file_write_iter(iocb, from, &cached);
2224 if (cached && result != -ENOSPC && result != -EDQUOT)
2225 GOTO(out, rc_normal = result);
2227 /* NB: we can't do direct IO for tiny writes because they use the page
2228 * cache, we can't do sync writes because tiny writes can't flush
2229 * pages, and we can't do append writes because we can't guarantee the
2230 * required DLM locks are held to protect file size.
2232 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2233 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
2234 rc_tiny = ll_do_tiny_write(iocb, from);
2236 /* In case of error, go on and try normal write - Only stop if tiny
2237 * write completed I/O.
2239 if (iov_iter_count(from) == 0)
2240 GOTO(out, rc_normal = rc_tiny);
2242 env = cl_env_get(&refcheck);
2244 RETURN(PTR_ERR(env));
2246 args = ll_env_args(env);
2247 args->u.normal.via_iter = from;
2248 args->u.normal.via_iocb = iocb;
2250 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2251 &iocb->ki_pos, iov_iter_count(from));
2253 /* On success, combine bytes written. */
2254 if (rc_tiny >= 0 && rc_normal > 0)
2255 rc_normal += rc_tiny;
2256 /* On error, only return error from normal write if tiny write did not
2257 * write any bytes. Otherwise return bytes written by tiny write.
2259 else if (rc_tiny > 0)
2260 rc_normal = rc_tiny;
2262 cl_env_put(env, &refcheck);
2264 if (rc_normal > 0) {
2265 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2266 file->private_data, iocb->ki_pos,
2268 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2269 ktime_us_delta(ktime_get(), kstart));
2273 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2274 file_dentry(file)->d_name.name,
2275 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2276 iov_iter_count(from));
2281 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2283 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2285 static int ll_file_get_iov_count(const struct iovec *iov,
2286 unsigned long *nr_segs, size_t *count,
2292 for (seg = 0; seg < *nr_segs; seg++) {
2293 const struct iovec *iv = &iov[seg];
2296 * If any segment has a negative length, or the cumulative
2297 * length ever wraps negative then return -EINVAL.
2300 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2302 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2307 cnt -= iv->iov_len; /* This segment is no good */
2314 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2315 unsigned long nr_segs, loff_t pos)
2322 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2329 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2330 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2331 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2332 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2333 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2335 result = ll_file_read_iter(iocb, &to);
2340 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2343 struct iovec iov = { .iov_base = buf, .iov_len = count };
2352 init_sync_kiocb(&kiocb, file);
2353 kiocb.ki_pos = *ppos;
2354 #ifdef HAVE_KIOCB_KI_LEFT
2355 kiocb.ki_left = count;
2356 #elif defined(HAVE_KI_NBYTES)
2357 kiocb.i_nbytes = count;
2360 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2361 *ppos = kiocb.ki_pos;
2367 * Write to a file (through the page cache).
2370 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2371 unsigned long nr_segs, loff_t pos)
2373 struct iov_iter from;
2378 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2385 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2386 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2387 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2388 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2389 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2391 result = ll_file_write_iter(iocb, &from);
2396 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2397 size_t count, loff_t *ppos)
2399 struct iovec iov = { .iov_base = (void __user *)buf,
2409 init_sync_kiocb(&kiocb, file);
2410 kiocb.ki_pos = *ppos;
2411 #ifdef HAVE_KIOCB_KI_LEFT
2412 kiocb.ki_left = count;
2413 #elif defined(HAVE_KI_NBYTES)
2414 kiocb.ki_nbytes = count;
2417 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2418 *ppos = kiocb.ki_pos;
2422 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2424 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2425 __u64 flags, struct lov_user_md *lum, int lum_size)
2427 struct lookup_intent oit = {
2429 .it_flags = flags | MDS_OPEN_BY_FID,
2434 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2435 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2436 /* this code will only exist for big-endian systems */
2437 lustre_swab_lov_user_md(lum, 0);
2440 ll_inode_size_lock(inode);
2441 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2443 GOTO(out_unlock, rc);
2445 ll_release_openhandle(dentry, &oit);
2448 ll_inode_size_unlock(inode);
2449 ll_intent_release(&oit);
2454 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2455 struct lov_mds_md **lmmp, int *lmm_size,
2456 struct ptlrpc_request **request)
2458 struct ll_sb_info *sbi = ll_i2sbi(inode);
2459 struct mdt_body *body;
2460 struct lov_mds_md *lmm = NULL;
2461 struct ptlrpc_request *req = NULL;
2462 struct md_op_data *op_data;
2467 rc = ll_get_default_mdsize(sbi, &lmmsize);
2471 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2472 strlen(filename), lmmsize,
2473 LUSTRE_OPC_ANY, NULL);
2474 if (IS_ERR(op_data))
2475 RETURN(PTR_ERR(op_data));
2477 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2478 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2479 ll_finish_md_op_data(op_data);
2481 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2486 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2487 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2489 lmmsize = body->mbo_eadatasize;
2491 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2493 GOTO(out, rc = -ENODATA);
2495 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2496 LASSERT(lmm != NULL);
2498 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2499 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2500 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2501 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2502 GOTO(out, rc = -EPROTO);
2505 * This is coming from the MDS, so is probably in
2506 * little endian. We convert it to host endian before
2507 * passing it to userspace.
2509 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2510 int stripe_count = 0;
2512 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2513 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2514 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2515 if (le32_to_cpu(lmm->lmm_pattern) &
2516 LOV_PATTERN_F_RELEASED)
2518 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2520 /* if function called for directory - we should
2521 * avoid swab not existent lsm objects
2523 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2524 S_ISREG(body->mbo_mode))
2525 lustre_swab_lov_user_md_objects(
2526 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2528 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2529 S_ISREG(body->mbo_mode))
2530 lustre_swab_lov_user_md_objects(
2531 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2533 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2534 lustre_swab_lov_comp_md_v1(
2535 (struct lov_comp_md_v1 *)lmm);
2539 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2540 struct lov_comp_md_v1 *comp_v1 = NULL;
2541 struct lov_comp_md_entry_v1 *ent;
2542 struct lov_user_md_v1 *v1;
2546 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2547 /* Dump the striping information */
2548 for (; i < comp_v1->lcm_entry_count; i++) {
2549 ent = &comp_v1->lcm_entries[i];
2550 off = ent->lcme_offset;
2551 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2553 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2554 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2558 * Return valid stripe_count and stripe_size instead of 0 for
2559 * DoM files to avoid divide-by-zero for older userspace that
2560 * calls this ioctl, e.g. lustre ADIO driver.
2562 if (lmm->lmm_stripe_count == 0)
2563 lmm->lmm_stripe_count = 1;
2564 if (lmm->lmm_stripe_size == 0) {
2565 /* Since the first component of the file data is placed
2566 * on the MDT for faster access, the stripe_size of the
2567 * second one is always that applications which are
2570 if (lmm->lmm_pattern == LOV_PATTERN_MDT)
2571 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2573 i = comp_v1->lcm_entry_count > 1 ?
2574 comp_v1->lcm_entry_count - 1 : 0;
2575 ent = &comp_v1->lcm_entries[i];
2576 off = ent->lcme_offset;
2577 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2578 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2583 *lmm_size = lmmsize;
2588 static int ll_lov_setea(struct inode *inode, struct file *file,
2591 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2592 struct lov_user_md *lump;
2593 int lum_size = sizeof(*lump) + sizeof(struct lov_user_ost_data);
2597 if (!capable(CAP_SYS_ADMIN))
2600 OBD_ALLOC_LARGE(lump, lum_size);
2604 if (copy_from_user(lump, arg, lum_size))
2605 GOTO(out_lump, rc = -EFAULT);
2607 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2609 cl_lov_delay_create_clear(&file->f_flags);
2612 OBD_FREE_LARGE(lump, lum_size);
2616 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2623 env = cl_env_get(&refcheck);
2625 RETURN(PTR_ERR(env));
2627 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2628 cl_env_put(env, &refcheck);
2632 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2635 struct lov_user_md __user *lum = arg;
2636 struct lov_user_md *klum;
2638 __u64 flags = FMODE_WRITE;
2641 rc = ll_copy_user_md(lum, &klum);
2646 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2651 rc = put_user(0, &lum->lmm_stripe_count);
2655 rc = ll_layout_refresh(inode, &gen);
2659 rc = ll_file_getstripe(inode, arg, lum_size);
2660 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2661 ll_i2info(inode)->lli_clob) {
2662 struct iattr attr = { 0 };
2664 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2665 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2668 cl_lov_delay_create_clear(&file->f_flags);
2671 OBD_FREE_LARGE(klum, lum_size);
2677 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2679 struct ll_inode_info *lli = ll_i2info(inode);
2680 struct cl_object *obj = lli->lli_clob;
2681 struct ll_file_data *fd = file->private_data;
2682 struct ll_grouplock grouplock;
2687 CWARN("group id for group lock must not be 0\n");
2691 if (ll_file_nolock(file))
2692 RETURN(-EOPNOTSUPP);
2694 if (file->f_flags & O_NONBLOCK) {
2695 if (!mutex_trylock(&lli->lli_group_mutex))
2698 mutex_lock(&lli->lli_group_mutex);
2701 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2702 CWARN("group lock already existed with gid %lu\n",
2703 fd->fd_grouplock.lg_gid);
2704 GOTO(out, rc = -EINVAL);
2706 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2707 if (file->f_flags & O_NONBLOCK)
2708 GOTO(out, rc = -EAGAIN);
2709 mutex_unlock(&lli->lli_group_mutex);
2710 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2711 GOTO(retry, rc = 0);
2713 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2716 * XXX: group lock needs to protect all OST objects while PFL
2717 * can add new OST objects during the IO, so we'd instantiate
2718 * all OST objects before getting its group lock.
2723 struct cl_layout cl = {
2724 .cl_is_composite = false,
2726 struct lu_extent ext = {
2728 .e_end = OBD_OBJECT_EOF,
2731 env = cl_env_get(&refcheck);
2733 GOTO(out, rc = PTR_ERR(env));
2735 rc = cl_object_layout_get(env, obj, &cl);
2736 if (rc >= 0 && cl.cl_is_composite)
2737 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2740 cl_env_put(env, &refcheck);
2745 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2746 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2751 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2752 fd->fd_grouplock = grouplock;
2753 if (lli->lli_group_users == 0)
2754 lli->lli_group_gid = grouplock.lg_gid;
2755 lli->lli_group_users++;
2757 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2759 mutex_unlock(&lli->lli_group_mutex);
2764 static int ll_put_grouplock(struct inode *inode, struct file *file,
2767 struct ll_inode_info *lli = ll_i2info(inode);
2768 struct ll_file_data *fd = file->private_data;
2769 struct ll_grouplock grouplock;
2773 mutex_lock(&lli->lli_group_mutex);
2774 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2775 CWARN("no group lock held\n");
2776 GOTO(out, rc = -EINVAL);
2779 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2781 if (fd->fd_grouplock.lg_gid != arg) {
2782 CWARN("group lock %lu doesn't match current id %lu\n",
2783 arg, fd->fd_grouplock.lg_gid);
2784 GOTO(out, rc = -EINVAL);
2787 grouplock = fd->fd_grouplock;
2788 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2789 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2791 cl_put_grouplock(&grouplock);
2793 lli->lli_group_users--;
2794 if (lli->lli_group_users == 0) {
2795 lli->lli_group_gid = 0;
2796 wake_up_var(&lli->lli_group_users);
2798 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2801 mutex_unlock(&lli->lli_group_mutex);
2807 * Close inode open handle
2809 * \param dentry [in] dentry which contains the inode
2810 * \param it [in,out] intent which contains open info and result
2813 * \retval <0 failure
2815 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2817 struct inode *inode = dentry->d_inode;
2818 struct obd_client_handle *och;
2824 /* Root ? Do nothing. */
2825 if (is_root_inode(inode))
2828 /* No open handle to close? Move away */
2829 if (!it_disposition(it, DISP_OPEN_OPEN))
2832 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2834 OBD_ALLOC(och, sizeof(*och));
2836 GOTO(out, rc = -ENOMEM);
2838 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2842 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2844 /* this one is in place of ll_file_open */
2845 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2846 ptlrpc_req_finished(it->it_request);
2847 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2853 * Get size for inode for which FIEMAP mapping is requested.
2854 * Make the FIEMAP get_info call and returns the result.
2855 * \param fiemap kernel buffer to hold extens
2856 * \param num_bytes kernel buffer size
2858 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2864 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2867 /* Checks for fiemap flags */
2868 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2869 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2873 /* Check for FIEMAP_FLAG_SYNC */
2874 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2875 rc = filemap_fdatawrite(inode->i_mapping);
2880 env = cl_env_get(&refcheck);
2882 RETURN(PTR_ERR(env));
2884 if (i_size_read(inode) == 0) {
2885 rc = ll_glimpse_size(inode);
2890 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLPROJID;
2891 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2892 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2894 /* If filesize is 0, then there would be no objects for mapping */
2895 if (fmkey.lfik_oa.o_size == 0) {
2896 fiemap->fm_mapped_extents = 0;
2900 fmkey.lfik_fiemap = *fiemap;
2902 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2903 &fmkey, fiemap, &num_bytes);
2905 cl_env_put(env, &refcheck);
2909 static int fid2path_for_enc_file(struct inode *parent, char *gfpath,
2912 struct dentry *de = NULL, *de_parent = d_find_any_alias(parent);
2913 struct llcrypt_str lltr = LLTR_INIT(NULL, 0);
2914 struct llcrypt_str de_name;
2915 char *p, *ptr = gfpath;
2916 size_t len = 0, len_orig = 0;
2917 int enckey = -1, nameenc = -1;
2921 while ((p = strsep(&gfpath, "/")) != NULL) {
2929 len_orig = strlen(p);
2931 rc = sscanf(p, "["SFID"]", RFID(&fid));
2933 p = strchr(p, ']') + 1;
2939 if (!IS_ENCRYPTED(parent)) {
2940 if (gfpathlen < len + 1) {
2945 memmove(ptr, p, len);
2949 gfpathlen -= len + 1;
2953 /* From here, we know parent is encrypted */
2956 rc = llcrypt_prepare_readdir(parent);
2957 if (rc && rc != -ENOKEY) {
2964 if (llcrypt_has_encryption_key(parent))
2970 llcrypt_policy_has_filename_enc(parent);
2973 /* Even if names are not encrypted, we still need to call
2974 * ll_fname_disk_to_usr in order to decode names as they are
2975 * coming from the wire.
2977 rc = llcrypt_fname_alloc_buffer(parent, NAME_MAX + 1, &lltr);
2985 rc = ll_fname_disk_to_usr(parent, 0, 0, &de_name,
2988 llcrypt_fname_free_buffer(&lltr);
2992 lltr.name[lltr.len] = '\0';
2994 if (lltr.len <= len_orig && gfpathlen >= lltr.len + 1) {
2995 memcpy(ptr, lltr.name, lltr.len);
3000 gfpathlen -= lltr.len + 1;
3004 llcrypt_fname_free_buffer(&lltr);
3006 if (rc == -EOVERFLOW) {
3013 /* We reached the end of the string, which means
3014 * we are dealing with the last component in the path.
3015 * So save a useless lookup and exit.
3021 if (enckey == 0 || nameenc == 0)
3025 de = lookup_one_len(p, de_parent, len);
3026 inode_unlock(parent);
3027 if (IS_ERR_OR_NULL(de) || !de->d_inode) {
3033 parent = de->d_inode;
3040 if (!IS_ERR_OR_NULL(de))
3045 int __ll_fid2path(struct inode *inode, struct getinfo_fid2path *gfout,
3046 size_t outsize, __u32 pathlen_orig)
3048 struct obd_export *exp = ll_i2mdexp(inode);
3051 /* Append root FID after gfout to let MDT know the root FID so that
3052 * it can lookup the correct path, this is mainly for fileset.
3053 * old server without fileset mount support will ignore this.
3055 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
3057 /* Call mdc_iocontrol */
3058 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
3060 if (!rc && gfout->gf_pathlen && gfout->gf_u.gf_path[0] == '/') {
3061 /* by convention, server side (mdt_path_current()) puts
3062 * a leading '/' to tell client that we are dealing with
3065 rc = fid2path_for_enc_file(inode, gfout->gf_u.gf_path,
3067 if (!rc && strlen(gfout->gf_u.gf_path) > pathlen_orig)
3074 int ll_fid2path(struct inode *inode, void __user *arg)
3076 const struct getinfo_fid2path __user *gfin = arg;
3077 __u32 pathlen, pathlen_orig;
3078 struct getinfo_fid2path *gfout;
3084 if (!capable(CAP_DAC_READ_SEARCH) &&
3085 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3088 /* Only need to get the buflen */
3089 if (get_user(pathlen, &gfin->gf_pathlen))
3092 if (pathlen > PATH_MAX)
3094 pathlen_orig = pathlen;
3097 outsize = sizeof(*gfout) + pathlen;
3098 OBD_ALLOC(gfout, outsize);
3102 if (copy_from_user(gfout, arg, sizeof(*gfout)))
3103 GOTO(gf_free, rc = -EFAULT);
3105 gfout->gf_pathlen = pathlen;
3106 rc = __ll_fid2path(inode, gfout, outsize, pathlen_orig);
3110 if (copy_to_user(arg, gfout, sizeof(*gfout) + pathlen_orig))
3114 OBD_FREE(gfout, outsize);
3115 if (rc == -ENAMETOOLONG) {
3116 pathlen += PATH_MAX;
3123 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
3125 struct cl_object *obj = ll_i2info(inode)->lli_clob;
3133 ioc->idv_version = 0;
3134 ioc->idv_layout_version = UINT_MAX;
3136 /* If no file object initialized, we consider its version is 0. */
3140 env = cl_env_get(&refcheck);
3142 RETURN(PTR_ERR(env));
3144 io = vvp_env_thread_io(env);
3146 io->u.ci_data_version.dv_data_version = 0;
3147 io->u.ci_data_version.dv_layout_version = UINT_MAX;
3148 io->u.ci_data_version.dv_flags = ioc->idv_flags;
3151 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
3152 result = cl_io_loop(env, io);
3154 result = io->ci_result;
3156 ioc->idv_version = io->u.ci_data_version.dv_data_version;
3157 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
3159 cl_io_fini(env, io);
3161 if (unlikely(io->ci_need_restart))
3164 cl_env_put(env, &refcheck);
3170 * Read the data_version for inode.
3172 * This value is computed using stripe object version on OST.
3173 * Version is computed using server side locking.
3175 * @param flags if do sync on the OST side;
3177 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
3178 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
3180 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
3182 struct ioc_data_version ioc = { .idv_flags = flags };
3185 rc = ll_ioc_data_version(inode, &ioc);
3187 *data_version = ioc.idv_version;
3193 * Trigger a HSM release request for the provided inode.
3195 int ll_hsm_release(struct inode *inode)
3198 struct obd_client_handle *och = NULL;
3199 __u64 data_version = 0;
3205 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
3206 ll_i2sbi(inode)->ll_fsname,
3207 PFID(&ll_i2info(inode)->lli_fid));
3209 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
3211 GOTO(out, rc = PTR_ERR(och));
3213 /* Grab latest data_version and [am]time values */
3214 rc = ll_data_version(inode, &data_version,
3215 LL_DV_WR_FLUSH | LL_DV_SZ_UPDATE);
3219 env = cl_env_get(&refcheck);
3221 GOTO(out, rc = PTR_ERR(env));
3223 rc = ll_merge_attr(env, inode);
3224 cl_env_put(env, &refcheck);
3226 /* If error happen, we have the wrong size for a file.
3232 /* Release the file.
3233 * NB: lease lock handle is released in mdc_hsm_release_pack() because
3234 * we still need it to pack l_remote_handle to MDT. */
3235 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
3241 if (och != NULL && !IS_ERR(och)) /* close the file */
3242 ll_lease_close(och, inode, NULL);
3247 struct ll_swap_stack {
3250 struct inode *inode1;
3251 struct inode *inode2;
3256 static int ll_swap_layouts(struct file *file1, struct file *file2,
3257 struct lustre_swap_layouts *lsl)
3259 struct mdc_swap_layouts msl;
3260 struct md_op_data *op_data;
3263 struct ll_swap_stack *llss = NULL;
3266 OBD_ALLOC_PTR(llss);
3270 llss->inode1 = file_inode(file1);
3271 llss->inode2 = file_inode(file2);
3273 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
3277 /* we use 2 bool because it is easier to swap than 2 bits */
3278 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
3279 llss->check_dv1 = true;
3281 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
3282 llss->check_dv2 = true;
3284 /* we cannot use lsl->sl_dvX directly because we may swap them */
3285 llss->dv1 = lsl->sl_dv1;
3286 llss->dv2 = lsl->sl_dv2;
3288 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
3289 if (rc == 0) /* same file, done! */
3292 if (rc < 0) { /* sequentialize it */
3293 swap(llss->inode1, llss->inode2);
3295 swap(llss->dv1, llss->dv2);
3296 swap(llss->check_dv1, llss->check_dv2);
3300 if (gid != 0) { /* application asks to flush dirty cache */
3301 rc = ll_get_grouplock(llss->inode1, file1, gid);
3305 rc = ll_get_grouplock(llss->inode2, file2, gid);
3307 ll_put_grouplock(llss->inode1, file1, gid);
3312 /* ultimate check, before swaping the layouts we check if
3313 * dataversion has changed (if requested) */
3314 if (llss->check_dv1) {
3315 rc = ll_data_version(llss->inode1, &dv, 0);
3318 if (dv != llss->dv1)
3319 GOTO(putgl, rc = -EAGAIN);
3322 if (llss->check_dv2) {
3323 rc = ll_data_version(llss->inode2, &dv, 0);
3326 if (dv != llss->dv2)
3327 GOTO(putgl, rc = -EAGAIN);
3330 /* struct md_op_data is used to send the swap args to the mdt
3331 * only flags is missing, so we use struct mdc_swap_layouts
3332 * through the md_op_data->op_data */
3333 /* flags from user space have to be converted before they are send to
3334 * server, no flag is sent today, they are only used on the client */
3337 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3338 0, LUSTRE_OPC_ANY, &msl);
3339 if (IS_ERR(op_data))
3340 GOTO(free, rc = PTR_ERR(op_data));
3342 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3343 sizeof(*op_data), op_data, NULL);
3344 ll_finish_md_op_data(op_data);
3351 ll_put_grouplock(llss->inode2, file2, gid);
3352 ll_put_grouplock(llss->inode1, file1, gid);
3362 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3364 struct obd_export *exp = ll_i2mdexp(inode);
3365 struct md_op_data *op_data;
3369 /* Detect out-of range masks */
3370 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3373 /* Non-root users are forbidden to set or clear flags which are
3374 * NOT defined in HSM_USER_MASK. */
3375 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3376 !capable(CAP_SYS_ADMIN))
3379 if (!exp_connect_archive_id_array(exp)) {
3380 /* Detect out-of range archive id */
3381 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3382 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3386 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3387 LUSTRE_OPC_ANY, hss);
3388 if (IS_ERR(op_data))
3389 RETURN(PTR_ERR(op_data));
3391 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3394 ll_finish_md_op_data(op_data);
3399 static int ll_hsm_import(struct inode *inode, struct file *file,
3400 struct hsm_user_import *hui)
3402 struct hsm_state_set *hss = NULL;
3403 struct iattr *attr = NULL;
3407 if (!S_ISREG(inode->i_mode))
3413 GOTO(out, rc = -ENOMEM);
3415 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3416 hss->hss_archive_id = hui->hui_archive_id;
3417 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3418 rc = ll_hsm_state_set(inode, hss);
3422 OBD_ALLOC_PTR(attr);
3424 GOTO(out, rc = -ENOMEM);
3426 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3427 attr->ia_mode |= S_IFREG;
3428 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3429 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3430 attr->ia_size = hui->hui_size;
3431 attr->ia_mtime.tv_sec = hui->hui_mtime;
3432 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3433 attr->ia_atime.tv_sec = hui->hui_atime;
3434 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3436 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3437 ATTR_UID | ATTR_GID |
3438 ATTR_MTIME | ATTR_MTIME_SET |
3439 ATTR_ATIME | ATTR_ATIME_SET;
3443 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3447 inode_unlock(inode);
3459 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3461 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3462 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3465 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3467 struct inode *inode = file_inode(file);
3469 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3470 ATTR_MTIME | ATTR_MTIME_SET |
3473 .tv_sec = lfu->lfu_atime_sec,
3474 .tv_nsec = lfu->lfu_atime_nsec,
3477 .tv_sec = lfu->lfu_mtime_sec,
3478 .tv_nsec = lfu->lfu_mtime_nsec,
3481 .tv_sec = lfu->lfu_ctime_sec,
3482 .tv_nsec = lfu->lfu_ctime_nsec,
3488 if (!capable(CAP_SYS_ADMIN))
3491 if (!S_ISREG(inode->i_mode))
3495 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3497 inode_unlock(inode);
3502 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3505 case MODE_READ_USER:
3507 case MODE_WRITE_USER:
3514 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3516 /* Used to allow the upper layers of the client to request an LDLM lock
3517 * without doing an actual read or write.
3519 * Used for ladvise lockahead to manually request specific locks.
3521 * \param[in] file file this ladvise lock request is on
3522 * \param[in] ladvise ladvise struct describing this lock request
3524 * \retval 0 success, no detailed result available (sync requests
3525 * and requests sent to the server [not handled locally]
3526 * cannot return detailed results)
3527 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3528 * see definitions for details.
3529 * \retval negative negative errno on error
3531 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3533 struct lu_env *env = NULL;
3534 struct cl_io *io = NULL;
3535 struct cl_lock *lock = NULL;
3536 struct cl_lock_descr *descr = NULL;
3537 struct dentry *dentry = file->f_path.dentry;
3538 struct inode *inode = dentry->d_inode;
3539 enum cl_lock_mode cl_mode;
3540 off_t start = ladvise->lla_start;
3541 off_t end = ladvise->lla_end;
3548 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3549 dentry, dentry->d_inode,
3550 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3553 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3555 GOTO(out, result = cl_mode);
3557 /* Get IO environment */
3558 result = cl_io_get(inode, &env, &io, &refcheck);
3562 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3565 * nothing to do for this io. This currently happens when
3566 * stripe sub-object's are not yet created.
3568 result = io->ci_result;
3569 } else if (result == 0) {
3570 lock = vvp_env_lock(env);
3571 descr = &lock->cll_descr;
3573 descr->cld_obj = io->ci_obj;
3574 /* Convert byte offsets to pages */
3575 descr->cld_start = cl_index(io->ci_obj, start);
3576 descr->cld_end = cl_index(io->ci_obj, end);
3577 descr->cld_mode = cl_mode;
3578 /* CEF_MUST is used because we do not want to convert a
3579 * lockahead request to a lockless lock */
3580 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3582 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3583 descr->cld_enq_flags |= CEF_SPECULATIVE;
3585 result = cl_lock_request(env, io, lock);
3587 /* On success, we need to release the lock */
3589 cl_lock_release(env, lock);
3591 cl_io_fini(env, io);
3592 cl_env_put(env, &refcheck);
3594 /* -ECANCELED indicates a matching lock with a different extent
3595 * was already present, and -EEXIST indicates a matching lock
3596 * on exactly the same extent was already present.
3597 * We convert them to positive values for userspace to make
3598 * recognizing true errors easier.
3599 * Note we can only return these detailed results on async requests,
3600 * as sync requests look the same as i/o requests for locking. */
3601 if (result == -ECANCELED)
3602 result = LLA_RESULT_DIFFERENT;
3603 else if (result == -EEXIST)
3604 result = LLA_RESULT_SAME;
3609 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3611 static int ll_ladvise_sanity(struct inode *inode,
3612 struct llapi_lu_ladvise *ladvise)
3614 struct ll_sb_info *sbi = ll_i2sbi(inode);
3615 enum lu_ladvise_type advice = ladvise->lla_advice;
3616 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3617 * be in the first 32 bits of enum ladvise_flags */
3618 __u32 flags = ladvise->lla_peradvice_flags;
3619 /* 3 lines at 80 characters per line, should be plenty */
3622 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3625 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3626 sbi->ll_fsname, advice,
3627 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3631 /* Per-advice checks */
3633 case LU_LADVISE_LOCKNOEXPAND:
3634 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3636 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3637 "rc = %d\n", sbi->ll_fsname, flags,
3638 ladvise_names[advice], rc);
3642 case LU_LADVISE_LOCKAHEAD:
3643 /* Currently only READ and WRITE modes can be requested */
3644 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3645 ladvise->lla_lockahead_mode == 0) {
3647 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3648 "rc = %d\n", sbi->ll_fsname,
3649 ladvise->lla_lockahead_mode,
3650 ladvise_names[advice], rc);
3654 case LU_LADVISE_WILLREAD:
3655 case LU_LADVISE_DONTNEED:
3657 /* Note fall through above - These checks apply to all advices
3658 * except LOCKNOEXPAND */
3659 if (flags & ~LF_DEFAULT_MASK) {
3661 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3662 "rc = %d\n", sbi->ll_fsname, flags,
3663 ladvise_names[advice], rc);
3666 if (ladvise->lla_start >= ladvise->lla_end) {
3668 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3669 "for %s: rc = %d\n", sbi->ll_fsname,
3670 ladvise->lla_start, ladvise->lla_end,
3671 ladvise_names[advice], rc);
3683 * Give file access advices
3685 * The ladvise interface is similar to Linux fadvise() system call, except it
3686 * forwards the advices directly from Lustre client to server. The server side
3687 * codes will apply appropriate read-ahead and caching techniques for the
3688 * corresponding files.
3690 * A typical workload for ladvise is e.g. a bunch of different clients are
3691 * doing small random reads of a file, so prefetching pages into OSS cache
3692 * with big linear reads before the random IO is a net benefit. Fetching
3693 * all that data into each client cache with fadvise() may not be, due to
3694 * much more data being sent to the client.
3696 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3697 struct llapi_lu_ladvise *ladvise)
3701 struct cl_ladvise_io *lio;
3706 env = cl_env_get(&refcheck);
3708 RETURN(PTR_ERR(env));
3710 io = vvp_env_thread_io(env);
3711 io->ci_obj = ll_i2info(inode)->lli_clob;
3713 /* initialize parameters for ladvise */
3714 lio = &io->u.ci_ladvise;
3715 lio->li_start = ladvise->lla_start;
3716 lio->li_end = ladvise->lla_end;
3717 lio->li_fid = ll_inode2fid(inode);
3718 lio->li_advice = ladvise->lla_advice;
3719 lio->li_flags = flags;
3721 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3722 rc = cl_io_loop(env, io);
3726 cl_io_fini(env, io);
3727 cl_env_put(env, &refcheck);
3731 static int ll_lock_noexpand(struct file *file, int flags)
3733 struct ll_file_data *fd = file->private_data;
3735 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3740 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3743 struct fsxattr fsxattr;
3745 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3748 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3749 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3750 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3751 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3752 if (copy_to_user(uarg, &fsxattr, sizeof(fsxattr)))
3758 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3762 * Project Quota ID state is only allowed to change from within the init
3763 * namespace. Enforce that restriction only if we are trying to change
3764 * the quota ID state. Everything else is allowed in user namespaces.
3766 if (current_user_ns() == &init_user_ns) {
3768 * Caller is allowed to change the project ID. if it is being
3769 * changed, make sure that the new value is valid.
3771 if (ll_i2info(inode)->lli_projid != projid &&
3772 !projid_valid(make_kprojid(&init_user_ns, projid)))
3778 if (ll_i2info(inode)->lli_projid != projid)
3781 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3782 if (!(xflags & FS_XFLAG_PROJINHERIT))
3785 if (xflags & FS_XFLAG_PROJINHERIT)
3792 static int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3794 struct ptlrpc_request *req = NULL;
3795 struct md_op_data *op_data;
3796 struct cl_object *obj;
3797 unsigned int inode_flags;
3800 CDEBUG(D_QUOTA, DFID" xflags=%x projid=%u\n",
3801 PFID(ll_inode2fid(inode)), xflags, projid);
3802 rc = ll_ioctl_check_project(inode, xflags, projid);
3806 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3807 LUSTRE_OPC_ANY, NULL);
3808 if (IS_ERR(op_data))
3809 RETURN(PTR_ERR(op_data));
3811 inode_flags = ll_xflags_to_inode_flags(xflags);
3812 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3813 if (xflags & FS_XFLAG_PROJINHERIT)
3814 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3816 /* pass projid to md_op_data */
3817 op_data->op_projid = projid;
3819 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3820 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3821 ptlrpc_req_finished(req);
3823 GOTO(out_fsxattr, rc);
3824 ll_update_inode_flags(inode, op_data->op_attr_flags);
3826 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3827 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3828 GOTO(out_fsxattr, rc);
3830 obj = ll_i2info(inode)->lli_clob;
3832 struct iattr attr = { 0 };
3834 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3838 ll_finish_md_op_data(op_data);
3842 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3845 struct fsxattr fsxattr;
3849 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3852 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3853 fsxattr.fsx_projid));
3856 int ll_ioctl_project(struct file *file, unsigned int cmd, void __user *uarg)
3858 struct lu_project lu_project;
3859 struct dentry *dentry = file_dentry(file);
3860 struct inode *inode = file_inode(file);
3861 struct dentry *child_dentry = NULL;
3862 int rc = 0, name_len;
3864 if (copy_from_user(&lu_project, uarg, sizeof(lu_project)))
3867 /* apply child dentry if name is valid */
3868 name_len = strnlen(lu_project.project_name, NAME_MAX);
3869 if (name_len > 0 && name_len <= NAME_MAX) {
3871 child_dentry = lookup_one_len(lu_project.project_name,
3873 inode_unlock(inode);
3874 if (IS_ERR(child_dentry)) {
3875 rc = PTR_ERR(child_dentry);
3878 inode = child_dentry->d_inode;
3883 } else if (name_len > NAME_MAX) {
3888 switch (lu_project.project_type) {
3889 case LU_PROJECT_SET:
3890 rc = ll_set_project(inode, lu_project.project_xflags,
3891 lu_project.project_id);
3893 case LU_PROJECT_GET:
3894 lu_project.project_xflags =
3895 ll_inode_flags_to_xflags(inode->i_flags);
3896 if (test_bit(LLIF_PROJECT_INHERIT,
3897 &ll_i2info(inode)->lli_flags))
3898 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3899 lu_project.project_id = ll_i2info(inode)->lli_projid;
3900 if (copy_to_user(uarg, &lu_project, sizeof(lu_project))) {
3910 if (!IS_ERR_OR_NULL(child_dentry))
3915 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3918 struct inode *inode = file_inode(file);
3919 struct ll_file_data *fd = file->private_data;
3920 struct ll_inode_info *lli = ll_i2info(inode);
3921 struct obd_client_handle *och = NULL;
3922 struct split_param sp;
3923 struct pcc_param param;
3924 bool lease_broken = false;
3926 enum mds_op_bias bias = 0;
3928 struct file *layout_file = NULL;
3930 size_t data_size = 0;
3931 bool attached = false;
3936 mutex_lock(&lli->lli_och_mutex);
3937 if (fd->fd_lease_och != NULL) {
3938 och = fd->fd_lease_och;
3939 fd->fd_lease_och = NULL;
3941 mutex_unlock(&lli->lli_och_mutex);
3946 fmode = och->och_flags;
3948 switch (ioc->lil_flags) {
3949 case LL_LEASE_RESYNC_DONE:
3950 if (ioc->lil_count > IOC_IDS_MAX)
3951 GOTO(out_lease_close, rc = -EINVAL);
3953 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3954 OBD_ALLOC(data, data_size);
3956 GOTO(out_lease_close, rc = -ENOMEM);
3958 if (copy_from_user(data, uarg, data_size))
3959 GOTO(out_lease_close, rc = -EFAULT);
3961 bias = MDS_CLOSE_RESYNC_DONE;
3963 case LL_LEASE_LAYOUT_MERGE: {
3964 if (ioc->lil_count != 1)
3965 GOTO(out_lease_close, rc = -EINVAL);
3967 uarg += sizeof(*ioc);
3968 if (copy_from_user(&fdv, uarg, sizeof(__u32)))
3969 GOTO(out_lease_close, rc = -EFAULT);
3971 layout_file = fget(fdv);
3973 GOTO(out_lease_close, rc = -EBADF);
3975 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3976 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3977 GOTO(out_lease_close, rc = -EPERM);
3979 data = file_inode(layout_file);
3980 bias = MDS_CLOSE_LAYOUT_MERGE;
3983 case LL_LEASE_LAYOUT_SPLIT: {
3986 if (ioc->lil_count != 2)
3987 GOTO(out_lease_close, rc = -EINVAL);
3989 uarg += sizeof(*ioc);
3990 if (copy_from_user(&fdv, uarg, sizeof(__u32)))
3991 GOTO(out_lease_close, rc = -EFAULT);
3993 uarg += sizeof(__u32);
3994 if (copy_from_user(&mirror_id, uarg, sizeof(__u32)))
3995 GOTO(out_lease_close, rc = -EFAULT);
3997 layout_file = fget(fdv);
3999 GOTO(out_lease_close, rc = -EBADF);
4001 /* if layout_file == file, it means to destroy the mirror */
4002 sp.sp_inode = file_inode(layout_file);
4003 sp.sp_mirror_id = (__u16)mirror_id;
4005 bias = MDS_CLOSE_LAYOUT_SPLIT;
4008 case LL_LEASE_PCC_ATTACH:
4009 if (ioc->lil_count != 1)
4012 if (IS_ENCRYPTED(inode))
4013 RETURN(-EOPNOTSUPP);
4015 uarg += sizeof(*ioc);
4016 if (copy_from_user(¶m.pa_archive_id, uarg, sizeof(__u32)))
4017 GOTO(out_lease_close, rc2 = -EFAULT);
4019 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
4021 GOTO(out_lease_close, rc2);
4024 /* Grab latest data version */
4025 rc2 = ll_data_version(inode, ¶m.pa_data_version,
4028 GOTO(out_lease_close, rc2);
4031 bias = MDS_PCC_ATTACH;
4034 /* without close intent */
4039 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
4043 rc = ll_lease_och_release(inode, file);
4052 if (ioc->lil_flags == LL_LEASE_RESYNC_DONE && data)
4053 OBD_FREE(data, data_size);
4058 if (ioc->lil_flags == LL_LEASE_PCC_ATTACH) {
4061 rc = pcc_readwrite_attach_fini(file, inode,
4062 param.pa_layout_gen,
4067 ll_layout_refresh(inode, &fd->fd_layout_version);
4070 rc = ll_lease_type_from_fmode(fmode);
4074 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
4077 struct inode *inode = file_inode(file);
4078 struct ll_inode_info *lli = ll_i2info(inode);
4079 struct ll_file_data *fd = file->private_data;
4080 struct obd_client_handle *och = NULL;
4081 __u64 open_flags = 0;
4087 switch (ioc->lil_mode) {
4088 case LL_LEASE_WRLCK:
4089 if (!(file->f_mode & FMODE_WRITE))
4091 fmode = FMODE_WRITE;
4093 case LL_LEASE_RDLCK:
4094 if (!(file->f_mode & FMODE_READ))
4098 case LL_LEASE_UNLCK:
4099 RETURN(ll_file_unlock_lease(file, ioc, uarg));
4104 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
4106 /* apply for lease */
4107 if (ioc->lil_flags & LL_LEASE_RESYNC)
4108 open_flags = MDS_OPEN_RESYNC;
4109 och = ll_lease_open(inode, file, fmode, open_flags);
4111 RETURN(PTR_ERR(och));
4113 if (ioc->lil_flags & LL_LEASE_RESYNC) {
4114 rc = ll_lease_file_resync(och, inode, uarg);
4116 ll_lease_close(och, inode, NULL);
4119 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
4121 ll_lease_close(och, inode, NULL);
4127 mutex_lock(&lli->lli_och_mutex);
4128 if (fd->fd_lease_och == NULL) {
4129 fd->fd_lease_och = och;
4132 mutex_unlock(&lli->lli_och_mutex);
4134 /* impossible now that only excl is supported for now */
4135 ll_lease_close(och, inode, &lease_broken);
4141 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
4143 struct ll_inode_info *lli = ll_i2info(inode);
4144 struct ll_sb_info *sbi = ll_i2sbi(inode);
4145 __u64 now = ktime_get_real_seconds();
4148 spin_lock(&lli->lli_heat_lock);
4149 heat->lh_flags = lli->lli_heat_flags;
4150 for (i = 0; i < heat->lh_count; i++)
4151 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
4152 now, sbi->ll_heat_decay_weight,
4153 sbi->ll_heat_period_second);
4154 spin_unlock(&lli->lli_heat_lock);
4157 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
4159 struct ll_inode_info *lli = ll_i2info(inode);
4162 spin_lock(&lli->lli_heat_lock);
4163 if (flags & LU_HEAT_FLAG_CLEAR)
4164 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
4166 if (flags & LU_HEAT_FLAG_OFF)
4167 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
4169 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
4171 spin_unlock(&lli->lli_heat_lock);
4177 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4179 struct inode *inode = file_inode(file);
4180 struct ll_file_data *fd = file->private_data;
4181 void __user *uarg = (void __user *)arg;
4185 CDEBUG(D_VFSTRACE|D_IOCTL, "VFS Op:inode="DFID"(%pK) cmd=%x arg=%lx\n",
4186 PFID(ll_inode2fid(inode)), inode, cmd, arg);
4187 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
4189 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
4190 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
4194 case LL_IOC_GETFLAGS:
4195 /* Get the current value of the file flags */
4196 return put_user(fd->fd_flags, (int __user *)arg);
4197 case LL_IOC_SETFLAGS:
4198 case LL_IOC_CLRFLAGS:
4199 /* Set or clear specific file flags */
4200 /* XXX This probably needs checks to ensure the flags are
4201 * not abused, and to handle any flag side effects.
4203 if (get_user(flags, (int __user *)arg))
4206 if (cmd == LL_IOC_SETFLAGS) {
4207 if ((flags & LL_FILE_IGNORE_LOCK) &&
4208 !(file->f_flags & O_DIRECT)) {
4210 CERROR("%s: unable to disable locking on non-O_DIRECT file "DFID": rc = %d\n",
4211 current->comm, PFID(ll_inode2fid(inode)),
4216 fd->fd_flags |= flags;
4218 fd->fd_flags &= ~flags;
4221 case LL_IOC_LOV_SETSTRIPE:
4222 case LL_IOC_LOV_SETSTRIPE_NEW:
4223 RETURN(ll_lov_setstripe(inode, file, uarg));
4224 case LL_IOC_LOV_SETEA:
4225 RETURN(ll_lov_setea(inode, file, uarg));
4226 case LL_IOC_LOV_SWAP_LAYOUTS: {
4228 struct lustre_swap_layouts lsl;
4230 if (copy_from_user(&lsl, uarg, sizeof(lsl)))
4233 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
4236 file2 = fget(lsl.sl_fd);
4240 /* O_WRONLY or O_RDWR */
4241 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
4242 GOTO(out, rc = -EPERM);
4244 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
4245 struct obd_client_handle *och = NULL;
4246 struct ll_inode_info *lli;
4247 struct inode *inode2;
4249 lli = ll_i2info(inode);
4250 mutex_lock(&lli->lli_och_mutex);
4251 if (fd->fd_lease_och != NULL) {
4252 och = fd->fd_lease_och;
4253 fd->fd_lease_och = NULL;
4255 mutex_unlock(&lli->lli_och_mutex);
4257 GOTO(out, rc = -ENOLCK);
4258 inode2 = file_inode(file2);
4259 rc = ll_swap_layouts_close(och, inode, inode2);
4261 rc = ll_swap_layouts(file, file2, &lsl);
4267 case LL_IOC_LOV_GETSTRIPE:
4268 case LL_IOC_LOV_GETSTRIPE_NEW:
4269 RETURN(ll_file_getstripe(inode, uarg, 0));
4270 case LL_IOC_GROUP_LOCK:
4271 RETURN(ll_get_grouplock(inode, file, arg));
4272 case LL_IOC_GROUP_UNLOCK:
4273 RETURN(ll_put_grouplock(inode, file, arg));
4274 case LL_IOC_DATA_VERSION: {
4275 struct ioc_data_version idv;
4278 if (copy_from_user(&idv, uarg, sizeof(idv)))
4281 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
4282 rc = ll_ioc_data_version(inode, &idv);
4284 if (rc == 0 && copy_to_user(uarg, &idv, sizeof(idv)))
4289 case LL_IOC_HSM_STATE_GET: {
4290 struct md_op_data *op_data;
4291 struct hsm_user_state *hus;
4298 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4299 LUSTRE_OPC_ANY, hus);
4300 if (IS_ERR(op_data)) {
4302 RETURN(PTR_ERR(op_data));
4305 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4308 if (copy_to_user(uarg, hus, sizeof(*hus)))
4311 ll_finish_md_op_data(op_data);
4315 case LL_IOC_HSM_STATE_SET: {
4316 struct hsm_state_set *hss;
4323 if (copy_from_user(hss, uarg, sizeof(*hss))) {
4328 rc = ll_hsm_state_set(inode, hss);
4333 case LL_IOC_HSM_ACTION: {
4334 struct md_op_data *op_data;
4335 struct hsm_current_action *hca;
4343 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4344 LUSTRE_OPC_ANY, hca);
4345 if (IS_ERR(op_data)) {
4347 RETURN(PTR_ERR(op_data));
4350 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4353 GOTO(skip_copy, rc);
4355 /* The hsm_current_action retreived from the server could
4356 * contain corrupt information. If it is incorrect data collect
4357 * debug information. We still send the data even if incorrect
4358 * to user land to handle.
4360 action = hsm_user_action2name(hca->hca_action);
4361 if (strcmp(action, "UNKNOWN") == 0 ||
4362 hca->hca_state > HPS_DONE) {
4364 "HSM current state %s action %s, offset = %llu, length %llu\n",
4365 hsm_progress_state2name(hca->hca_state), action,
4366 hca->hca_location.offset, hca->hca_location.length);
4369 if (copy_to_user(uarg, hca, sizeof(*hca)))
4372 ll_finish_md_op_data(op_data);
4376 case LL_IOC_SET_LEASE_OLD: {
4377 struct ll_ioc_lease ioc = { .lil_mode = arg };
4379 RETURN(ll_file_set_lease(file, &ioc, 0));
4381 case LL_IOC_SET_LEASE: {
4382 struct ll_ioc_lease ioc;
4384 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
4387 RETURN(ll_file_set_lease(file, &ioc, uarg));
4389 case LL_IOC_GET_LEASE: {
4390 struct ll_inode_info *lli = ll_i2info(inode);
4391 struct ldlm_lock *lock = NULL;
4394 mutex_lock(&lli->lli_och_mutex);
4395 if (fd->fd_lease_och != NULL) {
4396 struct obd_client_handle *och = fd->fd_lease_och;
4398 lock = ldlm_handle2lock(&och->och_lease_handle);
4400 lock_res_and_lock(lock);
4401 if (!ldlm_is_cancel(lock))
4402 fmode = och->och_flags;
4404 unlock_res_and_lock(lock);
4405 LDLM_LOCK_PUT(lock);
4408 mutex_unlock(&lli->lli_och_mutex);
4410 RETURN(ll_lease_type_from_fmode(fmode));
4412 case LL_IOC_HSM_IMPORT: {
4413 struct hsm_user_import *hui;
4419 if (copy_from_user(hui, uarg, sizeof(*hui))) {
4424 rc = ll_hsm_import(inode, file, hui);
4429 case LL_IOC_FUTIMES_3: {
4430 struct ll_futimes_3 lfu;
4432 if (copy_from_user(&lfu, uarg, sizeof(lfu)))
4435 RETURN(ll_file_futimes_3(file, &lfu));
4437 case LL_IOC_LADVISE: {
4438 struct llapi_ladvise_hdr *k_ladvise_hdr;
4439 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4442 int alloc_size = sizeof(*k_ladvise_hdr);
4445 u_ladvise_hdr = uarg;
4446 OBD_ALLOC_PTR(k_ladvise_hdr);
4447 if (k_ladvise_hdr == NULL)
4450 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4451 GOTO(out_ladvise, rc = -EFAULT);
4453 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4454 k_ladvise_hdr->lah_count < 1)
4455 GOTO(out_ladvise, rc = -EINVAL);
4457 num_advise = k_ladvise_hdr->lah_count;
4458 if (num_advise >= LAH_COUNT_MAX)
4459 GOTO(out_ladvise, rc = -EFBIG);
4461 OBD_FREE_PTR(k_ladvise_hdr);
4462 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4463 lah_advise[num_advise]);
4464 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4465 if (k_ladvise_hdr == NULL)
4469 * TODO: submit multiple advices to one server in a single RPC
4471 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4472 GOTO(out_ladvise, rc = -EFAULT);
4474 for (i = 0; i < num_advise; i++) {
4475 struct llapi_lu_ladvise *k_ladvise =
4476 &k_ladvise_hdr->lah_advise[i];
4477 struct llapi_lu_ladvise __user *u_ladvise =
4478 &u_ladvise_hdr->lah_advise[i];
4480 rc = ll_ladvise_sanity(inode, k_ladvise);
4482 GOTO(out_ladvise, rc);
4484 switch (k_ladvise->lla_advice) {
4485 case LU_LADVISE_LOCKNOEXPAND:
4486 rc = ll_lock_noexpand(file,
4487 k_ladvise->lla_peradvice_flags);
4488 GOTO(out_ladvise, rc);
4489 case LU_LADVISE_LOCKAHEAD:
4491 rc = ll_file_lock_ahead(file, k_ladvise);
4494 GOTO(out_ladvise, rc);
4497 &u_ladvise->lla_lockahead_result))
4498 GOTO(out_ladvise, rc = -EFAULT);
4501 rc = ll_ladvise(inode, file,
4502 k_ladvise_hdr->lah_flags,
4505 GOTO(out_ladvise, rc);
4512 OBD_FREE(k_ladvise_hdr, alloc_size);
4515 case LL_IOC_FLR_SET_MIRROR: {
4516 /* mirror I/O must be direct to avoid polluting page cache
4518 if (!(file->f_flags & O_DIRECT))
4521 fd->fd_designated_mirror = arg;
4524 case LL_IOC_HEAT_GET: {
4525 struct lu_heat uheat;
4526 struct lu_heat *heat;
4529 if (copy_from_user(&uheat, uarg, sizeof(uheat)))
4532 if (uheat.lh_count > OBD_HEAT_COUNT)
4533 uheat.lh_count = OBD_HEAT_COUNT;
4535 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4536 OBD_ALLOC(heat, size);
4540 heat->lh_count = uheat.lh_count;
4541 ll_heat_get(inode, heat);
4542 rc = copy_to_user(uarg, heat, size);
4543 OBD_FREE(heat, size);
4544 RETURN(rc ? -EFAULT : 0);
4546 case LL_IOC_HEAT_SET: {
4549 if (copy_from_user(&flags, uarg, sizeof(flags)))
4552 rc = ll_heat_set(inode, flags);
4555 case LL_IOC_PCC_DETACH: {
4556 struct lu_pcc_detach *detach;
4558 OBD_ALLOC_PTR(detach);
4562 if (copy_from_user(detach, uarg, sizeof(*detach)))
4563 GOTO(out_detach_free, rc = -EFAULT);
4565 if (!S_ISREG(inode->i_mode))
4566 GOTO(out_detach_free, rc = -EINVAL);
4568 if (!inode_owner_or_capable(&init_user_ns, inode))
4569 GOTO(out_detach_free, rc = -EPERM);
4571 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4573 OBD_FREE_PTR(detach);
4576 case LL_IOC_PCC_STATE: {
4577 struct lu_pcc_state __user *ustate = uarg;
4578 struct lu_pcc_state *state;
4580 OBD_ALLOC_PTR(state);
4584 if (copy_from_user(state, ustate, sizeof(*state)))
4585 GOTO(out_state, rc = -EFAULT);
4587 rc = pcc_ioctl_state(file, inode, state);
4589 GOTO(out_state, rc);
4591 if (copy_to_user(ustate, state, sizeof(*state)))
4592 GOTO(out_state, rc = -EFAULT);
4595 OBD_FREE_PTR(state);
4599 rc = ll_iocontrol(inode, file, cmd, uarg);
4602 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, uarg));
4606 loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4608 struct inode *inode = file_inode(file);
4611 struct cl_lseek_io *lsio;
4618 env = cl_env_get(&refcheck);
4620 RETURN(PTR_ERR(env));
4622 io = vvp_env_thread_io(env);
4623 io->ci_obj = ll_i2info(inode)->lli_clob;
4624 ll_io_set_mirror(io, file);
4626 lsio = &io->u.ci_lseek;
4627 lsio->ls_start = offset;
4628 lsio->ls_whence = whence;
4629 lsio->ls_result = -ENXIO;
4632 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4634 struct vvp_io *vio = vvp_env_io(env);
4636 vio->vui_fd = file->private_data;
4637 rc = cl_io_loop(env, io);
4641 retval = rc ? : lsio->ls_result;
4642 cl_io_fini(env, io);
4643 } while (unlikely(io->ci_need_restart));
4645 cl_env_put(env, &refcheck);
4647 /* Without the key, SEEK_HOLE return value has to be
4648 * rounded up to next LUSTRE_ENCRYPTION_UNIT_SIZE.
4650 if (llcrypt_require_key(inode) == -ENOKEY && whence == SEEK_HOLE)
4651 retval = round_up(retval, LUSTRE_ENCRYPTION_UNIT_SIZE);
4656 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4658 struct inode *inode = file_inode(file);
4659 loff_t retval = offset, eof = 0;
4660 ktime_t kstart = ktime_get();
4664 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4665 PFID(ll_inode2fid(inode)), inode, retval, retval,
4668 if (origin == SEEK_END) {
4669 retval = ll_glimpse_size(inode);
4672 eof = i_size_read(inode);
4675 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4679 /* flush local cache first if any */
4680 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4683 retval = ll_lseek(file, offset, origin);
4686 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4688 retval = generic_file_llseek_size(file, offset, origin,
4689 ll_file_maxbytes(inode), eof);
4692 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4693 ktime_us_delta(ktime_get(), kstart));
4697 static int ll_flush(struct file *file, fl_owner_t id)
4699 struct inode *inode = file_inode(file);
4700 struct ll_inode_info *lli = ll_i2info(inode);
4701 struct ll_file_data *fd = file->private_data;
4704 LASSERT(!S_ISDIR(inode->i_mode));
4706 /* catch async errors that were recorded back when async writeback
4707 * failed for pages in this mapping. */
4708 rc = lli->lli_async_rc;
4709 lli->lli_async_rc = 0;
4710 if (lli->lli_clob != NULL) {
4711 err = lov_read_and_clear_async_rc(lli->lli_clob);
4716 /* The application has been told write failure already.
4717 * Do not report failure again. */
4718 if (fd->fd_write_failed)
4720 return rc ? -EIO : 0;
4724 * Called to make sure a portion of file has been written out.
4725 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4727 * Return how many pages have been written.
4729 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4730 enum cl_fsync_mode mode, int ignore_layout)
4734 struct cl_fsync_io *fio;
4739 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4740 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4743 env = cl_env_get(&refcheck);
4745 RETURN(PTR_ERR(env));
4747 io = vvp_env_thread_io(env);
4748 io->ci_obj = ll_i2info(inode)->lli_clob;
4749 io->ci_ignore_layout = ignore_layout;
4751 /* initialize parameters for sync */
4752 fio = &io->u.ci_fsync;
4753 fio->fi_start = start;
4755 fio->fi_fid = ll_inode2fid(inode);
4756 fio->fi_mode = mode;
4757 fio->fi_nr_written = 0;
4759 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4760 result = cl_io_loop(env, io);
4762 result = io->ci_result;
4764 result = fio->fi_nr_written;
4765 cl_io_fini(env, io);
4766 cl_env_put(env, &refcheck);
4772 * When dentry is provided (the 'else' case), file_dentry() may be
4773 * null and dentry must be used directly rather than pulled from
4774 * file_dentry() as is done otherwise.
4777 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4779 struct dentry *dentry = file_dentry(file);
4780 struct inode *inode = dentry->d_inode;
4781 struct ll_inode_info *lli = ll_i2info(inode);
4782 struct ptlrpc_request *req;
4783 ktime_t kstart = ktime_get();
4789 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4790 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4792 /* fsync's caller has already called _fdata{sync,write}, we want
4793 * that IO to finish before calling the osc and mdc sync methods */
4794 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4796 /* catch async errors that were recorded back when async writeback
4797 * failed for pages in this mapping. */
4798 if (!S_ISDIR(inode->i_mode)) {
4799 err = lli->lli_async_rc;
4800 lli->lli_async_rc = 0;
4803 if (lli->lli_clob != NULL) {
4804 err = lov_read_and_clear_async_rc(lli->lli_clob);
4810 if (S_ISREG(inode->i_mode) && !lli->lli_synced_to_mds) {
4812 * only the first sync on MDS makes sense,
4813 * everything else is stored on OSTs
4815 err = md_fsync(ll_i2sbi(inode)->ll_md_exp,
4816 ll_inode2fid(inode), &req);
4820 lli->lli_synced_to_mds = true;
4821 ptlrpc_req_finished(req);
4825 if (S_ISREG(inode->i_mode)) {
4826 struct ll_file_data *fd = file->private_data;
4829 /* Sync metadata on MDT first, and then sync the cached data
4832 err = pcc_fsync(file, start, end, datasync, &cached);
4834 err = cl_sync_file_range(inode, start, end,
4836 if (rc == 0 && err < 0)
4839 fd->fd_write_failed = true;
4841 fd->fd_write_failed = false;
4845 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4846 ktime_us_delta(ktime_get(), kstart));
4851 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4853 struct inode *inode = file_inode(file);
4854 struct ll_sb_info *sbi = ll_i2sbi(inode);
4855 struct ldlm_enqueue_info einfo = {
4856 .ei_type = LDLM_FLOCK,
4857 .ei_cb_cp = ldlm_flock_completion_ast,
4858 .ei_cbdata = file_lock,
4860 struct md_op_data *op_data;
4861 struct lustre_handle lockh = { 0 };
4862 union ldlm_policy_data flock = { { 0 } };
4863 int fl_type = file_lock->fl_type;
4864 ktime_t kstart = ktime_get();
4870 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4871 PFID(ll_inode2fid(inode)), file_lock);
4873 if (file_lock->fl_flags & FL_FLOCK) {
4874 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4875 /* flocks are whole-file locks */
4876 flock.l_flock.end = OFFSET_MAX;
4877 /* For flocks owner is determined by the local file desctiptor*/
4878 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4879 } else if (file_lock->fl_flags & FL_POSIX) {
4880 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4881 flock.l_flock.start = file_lock->fl_start;
4882 flock.l_flock.end = file_lock->fl_end;
4886 flock.l_flock.pid = file_lock->fl_pid;
4888 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4889 /* Somewhat ugly workaround for svc lockd.
4890 * lockd installs custom fl_lmops->lm_compare_owner that checks
4891 * for the fl_owner to be the same (which it always is on local node
4892 * I guess between lockd processes) and then compares pid.
4893 * As such we assign pid to the owner field to make it all work,
4894 * conflict with normal locks is unlikely since pid space and
4895 * pointer space for current->files are not intersecting */
4896 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4897 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4902 einfo.ei_mode = LCK_PR;
4905 /* An unlock request may or may not have any relation to
4906 * existing locks so we may not be able to pass a lock handle
4907 * via a normal ldlm_lock_cancel() request. The request may even
4908 * unlock a byte range in the middle of an existing lock. In
4909 * order to process an unlock request we need all of the same
4910 * information that is given with a normal read or write record
4911 * lock request. To avoid creating another ldlm unlock (cancel)
4912 * message we'll treat a LCK_NL flock request as an unlock. */
4913 einfo.ei_mode = LCK_NL;
4916 einfo.ei_mode = LCK_PW;
4920 CERROR("%s: fcntl from '%s' unknown lock type=%d: rc = %d\n",
4921 sbi->ll_fsname, current->comm, fl_type, rc);
4936 flags = LDLM_FL_BLOCK_NOWAIT;
4942 flags = LDLM_FL_TEST_LOCK;
4946 CERROR("%s: fcntl from '%s' unknown lock command=%d: rc = %d\n",
4947 sbi->ll_fsname, current->comm, cmd, rc);
4951 /* Save the old mode so that if the mode in the lock changes we
4952 * can decrement the appropriate reader or writer refcount. */
4953 file_lock->fl_type = einfo.ei_mode;
4955 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4956 LUSTRE_OPC_ANY, NULL);
4957 if (IS_ERR(op_data))
4958 RETURN(PTR_ERR(op_data));
4960 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4961 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4962 flock.l_flock.pid, flags, einfo.ei_mode,
4963 flock.l_flock.start, flock.l_flock.end);
4965 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4968 /* Restore the file lock type if not TEST lock. */
4969 if (!(flags & LDLM_FL_TEST_LOCK))
4970 file_lock->fl_type = fl_type;
4972 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4973 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4974 !(flags & LDLM_FL_TEST_LOCK))
4975 rc2 = locks_lock_file_wait(file, file_lock);
4977 if ((file_lock->fl_flags & FL_FLOCK) &&
4978 (rc == 0 || file_lock->fl_type == F_UNLCK))
4979 rc2 = flock_lock_file_wait(file, file_lock);
4980 if ((file_lock->fl_flags & FL_POSIX) &&
4981 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4982 !(flags & LDLM_FL_TEST_LOCK))
4983 rc2 = posix_lock_file_wait(file, file_lock);
4984 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4986 if (rc2 && file_lock->fl_type != F_UNLCK) {
4987 einfo.ei_mode = LCK_NL;
4988 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4993 ll_finish_md_op_data(op_data);
4996 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
4997 ktime_us_delta(ktime_get(), kstart));
5001 int ll_get_fid_by_name(struct inode *parent, const char *name,
5002 int namelen, struct lu_fid *fid,
5003 struct inode **inode)
5005 struct md_op_data *op_data = NULL;
5006 struct mdt_body *body;
5007 struct ptlrpc_request *req;
5011 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
5012 LUSTRE_OPC_ANY, NULL);
5013 if (IS_ERR(op_data))
5014 RETURN(PTR_ERR(op_data));
5016 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
5017 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
5018 ll_finish_md_op_data(op_data);
5022 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
5024 GOTO(out_req, rc = -EFAULT);
5026 *fid = body->mbo_fid1;
5029 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
5031 ptlrpc_req_finished(req);
5035 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
5036 const char *name, __u32 flags)
5038 struct dentry *dchild = NULL;
5039 struct inode *child_inode = NULL;
5040 struct md_op_data *op_data;
5041 struct ptlrpc_request *request = NULL;
5042 struct obd_client_handle *och = NULL;
5044 struct mdt_body *body;
5045 __u64 data_version = 0;
5046 size_t namelen = strlen(name);
5047 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
5051 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
5052 PFID(ll_inode2fid(parent)), name,
5053 lum->lum_stripe_offset, lum->lum_stripe_count);
5055 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
5056 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
5057 lustre_swab_lmv_user_md(lum);
5059 /* Get child FID first */
5060 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
5063 dchild = d_lookup(file_dentry(file), &qstr);
5065 if (dchild->d_inode)
5066 child_inode = igrab(dchild->d_inode);
5071 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
5080 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
5081 OBD_CONNECT2_DIR_MIGRATE)) {
5082 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
5083 ll_dir_striped(child_inode)) {
5084 CERROR("%s: MDT doesn't support stripe directory "
5085 "migration!\n", ll_i2sbi(parent)->ll_fsname);
5086 GOTO(out_iput, rc = -EOPNOTSUPP);
5091 * lfs migrate command needs to be blocked on the client
5092 * by checking the migrate FID against the FID of the
5095 if (is_root_inode(child_inode))
5096 GOTO(out_iput, rc = -EINVAL);
5098 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
5099 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
5100 if (IS_ERR(op_data))
5101 GOTO(out_iput, rc = PTR_ERR(op_data));
5103 inode_lock(child_inode);
5104 op_data->op_fid3 = *ll_inode2fid(child_inode);
5105 if (!fid_is_sane(&op_data->op_fid3)) {
5106 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
5107 ll_i2sbi(parent)->ll_fsname, name,
5108 PFID(&op_data->op_fid3));
5109 GOTO(out_unlock, rc = -EINVAL);
5112 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
5113 op_data->op_data = lum;
5114 op_data->op_data_size = lumlen;
5116 /* migrate dirent only for subdirs if MDS_MIGRATE_NSONLY set */
5117 if (S_ISDIR(child_inode->i_mode) && (flags & MDS_MIGRATE_NSONLY) &&
5118 lmv_dir_layout_changing(ll_i2info(parent)->lli_lsm_md))
5119 op_data->op_bias |= MDS_MIGRATE_NSONLY;
5122 if (S_ISREG(child_inode->i_mode)) {
5123 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
5127 GOTO(out_unlock, rc);
5130 rc = ll_data_version(child_inode, &data_version,
5133 GOTO(out_close, rc);
5135 op_data->op_open_handle = och->och_open_handle;
5136 op_data->op_data_version = data_version;
5137 op_data->op_lease_handle = och->och_lease_handle;
5138 op_data->op_bias |= MDS_CLOSE_MIGRATE;
5140 spin_lock(&och->och_mod->mod_open_req->rq_lock);
5141 och->och_mod->mod_open_req->rq_replay = 0;
5142 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
5145 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
5146 op_data->op_name, op_data->op_namelen,
5147 op_data->op_name, op_data->op_namelen, &request);
5149 LASSERT(request != NULL);
5150 ll_update_times(request, parent);
5153 if (rc == 0 || rc == -EAGAIN) {
5154 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
5155 LASSERT(body != NULL);
5157 /* If the server does release layout lock, then we cleanup
5158 * the client och here, otherwise release it in out_close: */
5159 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
5160 obd_mod_put(och->och_mod);
5161 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
5163 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
5169 if (request != NULL) {
5170 ptlrpc_req_finished(request);
5174 /* Try again if the lease has cancelled. */
5175 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
5180 ll_lease_close(och, child_inode, NULL);
5182 clear_nlink(child_inode);
5184 inode_unlock(child_inode);
5185 ll_finish_md_op_data(op_data);
5192 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
5194 struct ll_file_data *fd = file->private_data;
5198 * In order to avoid flood of warning messages, only print one message
5199 * for one file. And the entire message rate on the client is limited
5200 * by CDEBUG_LIMIT too.
5202 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
5203 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
5204 CDEBUG_LIMIT(D_CONSOLE,
5205 "flock disabled, mount with '-o [local]flock' to enable\r\n");
5211 * test if some locks matching bits and l_req_mode are acquired
5212 * - bits can be in different locks
5213 * - if found clear the common lock bits in *bits
5214 * - the bits not found, are kept in *bits
5216 * \param bits [IN] searched lock bits [IN]
5217 * \param l_req_mode [IN] searched lock mode
5218 * \retval boolean, true iff all bits are found
5220 int ll_have_md_lock(struct obd_export *exp, struct inode *inode, __u64 *bits,
5221 enum ldlm_mode l_req_mode)
5223 struct lustre_handle lockh;
5224 union ldlm_policy_data policy;
5225 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
5226 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5235 fid = &ll_i2info(inode)->lli_fid;
5236 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5237 ldlm_lockname[mode]);
5239 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5240 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5241 policy.l_inodebits.bits = *bits & BIT(i);
5242 if (policy.l_inodebits.bits == 0)
5245 if (md_lock_match(exp, flags, fid, LDLM_IBITS, &policy, mode,
5247 struct ldlm_lock *lock;
5249 lock = ldlm_handle2lock(&lockh);
5252 ~(lock->l_policy_data.l_inodebits.bits);
5253 LDLM_LOCK_PUT(lock);
5255 *bits &= ~policy.l_inodebits.bits;
5262 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5263 struct lustre_handle *lockh, __u64 flags,
5264 enum ldlm_mode mode)
5266 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5271 fid = &ll_i2info(inode)->lli_fid;
5272 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5274 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5275 fid, LDLM_IBITS, &policy, mode, lockh);
5280 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5282 /* Already unlinked. Just update nlink and return success */
5283 if (rc == -ENOENT) {
5285 /* If it is striped directory, and there is bad stripe
5286 * Let's revalidate the dentry again, instead of returning
5288 if (ll_dir_striped(inode))
5291 /* This path cannot be hit for regular files unless in
5292 * case of obscure races, so no need to to validate
5294 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5296 } else if (rc != 0) {
5297 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5298 "%s: revalidate FID "DFID" error: rc = %d\n",
5299 ll_i2sbi(inode)->ll_fsname,
5300 PFID(ll_inode2fid(inode)), rc);
5306 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5308 struct inode *parent;
5309 struct inode *inode = dentry->d_inode;
5310 struct obd_export *exp = ll_i2mdexp(inode);
5311 struct lookup_intent oit = {
5314 struct ptlrpc_request *req = NULL;
5315 struct md_op_data *op_data;
5316 const char *name = NULL;
5321 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5322 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5324 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5325 parent = dentry->d_parent->d_inode;
5326 name = dentry->d_name.name;
5327 namelen = dentry->d_name.len;
5332 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5333 LUSTRE_OPC_ANY, NULL);
5334 if (IS_ERR(op_data))
5335 RETURN(PTR_ERR(op_data));
5337 /* Call getattr by fid */
5338 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5339 op_data->op_flags = MF_GETATTR_BY_FID;
5340 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5341 ll_finish_md_op_data(op_data);
5343 rc = ll_inode_revalidate_fini(inode, rc);
5347 rc = ll_revalidate_it_finish(req, &oit, dentry);
5349 ll_intent_release(&oit);
5353 /* Unlinked? Unhash dentry, so it is not picked up later by
5354 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5355 * here to preserve get_cwd functionality on 2.6.
5357 if (!dentry->d_inode->i_nlink)
5358 d_lustre_invalidate(dentry);
5360 ll_lookup_finish_locks(&oit, dentry);
5362 ptlrpc_req_finished(req);
5367 static int ll_merge_md_attr(struct inode *inode)
5369 struct ll_inode_info *lli = ll_i2info(inode);
5370 struct cl_attr attr = { 0 };
5373 if (!lli->lli_lsm_md)
5376 down_read(&lli->lli_lsm_sem);
5377 if (!lmv_dir_striped(lli->lli_lsm_md)) {
5378 up_read(&lli->lli_lsm_sem);
5381 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
5382 &attr, ll_md_blocking_ast);
5383 up_read(&lli->lli_lsm_sem);
5387 spin_lock(&inode->i_lock);
5388 set_nlink(inode, attr.cat_nlink);
5389 spin_unlock(&inode->i_lock);
5391 inode->i_blocks = attr.cat_blocks;
5392 i_size_write(inode, attr.cat_size);
5394 ll_i2info(inode)->lli_atime = attr.cat_atime;
5395 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5396 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5401 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5402 unsigned int flags, bool foreign)
5404 struct inode *inode = de->d_inode;
5405 struct ll_sb_info *sbi = ll_i2sbi(inode);
5406 struct ll_inode_info *lli = ll_i2info(inode);
5407 struct inode *dir = de->d_parent->d_inode;
5408 bool need_glimpse = true;
5409 ktime_t kstart = ktime_get();
5412 /* The OST object(s) determine the file size, blocks and mtime. */
5413 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5414 request_mask & STATX_MTIME))
5415 need_glimpse = false;
5417 if (dentry_may_statahead(dir, de))
5418 ll_start_statahead(dir, de, need_glimpse &&
5419 !(flags & AT_STATX_DONT_SYNC));
5421 if (flags & AT_STATX_DONT_SYNC)
5422 GOTO(fill_attr, rc = 0);
5424 rc = ll_inode_revalidate(de, IT_GETATTR);
5428 /* foreign file/dir are always of zero length, so don't
5429 * need to validate size.
5431 if (S_ISREG(inode->i_mode) && !foreign) {
5435 GOTO(fill_attr, rc);
5437 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5438 if (cached && rc < 0)
5442 GOTO(fill_attr, rc);
5445 * If the returned attr is masked with OBD_MD_FLSIZE &
5446 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5447 * or blocks obtained from MDT is strictly correct, and the file
5448 * is usually not being modified by clients, and the [a|m|c]time
5449 * got from MDT is also strictly correct.
5450 * Under this circumstance, it does not need to send glimpse
5451 * RPCs to OSTs for file attributes such as the size and blocks.
5453 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5454 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5455 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5456 inode->i_mtime.tv_sec = lli->lli_mtime;
5457 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5458 inode->i_atime.tv_sec = lli->lli_atime;
5459 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5460 inode->i_ctime.tv_sec = lli->lli_ctime;
5461 GOTO(fill_attr, rc);
5464 /* In case of restore, the MDT has the right size and has
5465 * already send it back without granting the layout lock,
5466 * inode is up-to-date so glimpse is useless.
5467 * Also to glimpse we need the layout, in case of a running
5468 * restore the MDT holds the layout lock so the glimpse will
5469 * block up to the end of restore (getattr will block)
5471 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5472 rc = ll_glimpse_size(inode);
5477 /* If object isn't regular a file then don't validate size. */
5478 /* foreign dir is not striped dir */
5479 if (ll_dir_striped(inode) && !foreign) {
5480 rc = ll_merge_md_attr(inode);
5485 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5486 inode->i_atime.tv_sec = lli->lli_atime;
5487 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5488 inode->i_mtime.tv_sec = lli->lli_mtime;
5489 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5490 inode->i_ctime.tv_sec = lli->lli_ctime;
5494 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5496 if (ll_need_32bit_api(sbi)) {
5497 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5498 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5499 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5501 stat->ino = inode->i_ino;
5502 stat->dev = inode->i_sb->s_dev;
5503 stat->rdev = inode->i_rdev;
5506 /* foreign symlink to be exposed as a real symlink */
5508 stat->mode = inode->i_mode;
5510 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5512 stat->uid = inode->i_uid;
5513 stat->gid = inode->i_gid;
5514 stat->atime = inode->i_atime;
5515 stat->mtime = inode->i_mtime;
5516 stat->ctime = inode->i_ctime;
5517 /* stat->blksize is used to tell about preferred IO size */
5518 if (sbi->ll_stat_blksize)
5519 stat->blksize = sbi->ll_stat_blksize;
5520 else if (S_ISREG(inode->i_mode))
5521 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
5522 LL_MAX_BLKSIZE_BITS);
5524 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5526 stat->nlink = inode->i_nlink;
5527 stat->size = i_size_read(inode);
5528 stat->blocks = inode->i_blocks;
5530 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5531 if (flags & AT_STATX_DONT_SYNC) {
5532 if (stat->size == 0 &&
5533 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5534 stat->size = lli->lli_lazysize;
5535 if (stat->blocks == 0 &&
5536 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5537 stat->blocks = lli->lli_lazyblocks;
5540 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5541 stat->result_mask |= STATX_BTIME;
5542 stat->btime.tv_sec = lli->lli_btime;
5545 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5546 #ifdef HAVE_LUSTRE_CRYPTO
5547 stat->attributes_mask |= STATX_ATTR_ENCRYPTED;
5549 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5550 /* if Lustre specific LUSTRE_ENCRYPT_FL flag is set, also set
5551 * ext4 equivalent to please statx
5553 if (stat->attributes & LUSTRE_ENCRYPT_FL)
5554 stat->attributes |= STATX_ATTR_ENCRYPTED;
5555 stat->result_mask &= request_mask;
5558 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5559 ktime_us_delta(ktime_get(), kstart));
5564 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5565 int ll_getattr(struct user_namespace *mnt_userns, const struct path *path,
5566 struct kstat *stat, u32 request_mask, unsigned int flags)
5568 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5572 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5574 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5575 AT_STATX_SYNC_AS_STAT, false);
5579 int cl_falloc(struct file *file, struct inode *inode, int mode, loff_t offset,
5582 loff_t size = i_size_read(inode);
5590 env = cl_env_get(&refcheck);
5592 RETURN(PTR_ERR(env));
5594 io = vvp_env_thread_io(env);
5595 io->ci_obj = ll_i2info(inode)->lli_clob;
5596 ll_io_set_mirror(io, file);
5598 io->ci_verify_layout = 1;
5599 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5600 io->u.ci_setattr.sa_falloc_mode = mode;
5601 io->u.ci_setattr.sa_falloc_offset = offset;
5602 io->u.ci_setattr.sa_falloc_end = offset + len;
5603 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5605 CDEBUG(D_INODE, "UID %u GID %u PRJID %u\n",
5606 from_kuid(&init_user_ns, inode->i_uid),
5607 from_kgid(&init_user_ns, inode->i_gid),
5608 ll_i2info(inode)->lli_projid);
5610 io->u.ci_setattr.sa_falloc_uid = from_kuid(&init_user_ns, inode->i_uid);
5611 io->u.ci_setattr.sa_falloc_gid = from_kgid(&init_user_ns, inode->i_gid);
5612 io->u.ci_setattr.sa_falloc_projid = ll_i2info(inode)->lli_projid;
5614 if (io->u.ci_setattr.sa_falloc_end > size) {
5615 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5617 /* Check new size against VFS/VM file size limit and rlimit */
5618 rc = inode_newsize_ok(inode, newsize);
5621 if (newsize > ll_file_maxbytes(inode)) {
5622 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5623 (unsigned long long)newsize,
5624 ll_file_maxbytes(inode));
5631 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5633 rc = cl_io_loop(env, io);
5636 cl_io_fini(env, io);
5637 } while (unlikely(io->ci_need_restart));
5640 cl_env_put(env, &refcheck);
5644 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5646 struct inode *inode = file_inode(filp);
5649 if (offset < 0 || len <= 0)
5652 * Encrypted inodes can't handle collapse range or zero range or insert
5653 * range since we would need to re-encrypt blocks with a different IV or
5654 * XTS tweak (which are based on the logical block number).
5655 * Similar to what ext4 does.
5657 if (IS_ENCRYPTED(inode) &&
5658 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5659 FALLOC_FL_ZERO_RANGE)))
5660 RETURN(-EOPNOTSUPP);
5663 * mode == 0 (which is standard prealloc) and PUNCH is supported
5664 * Rest of mode options are not supported yet.
5666 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5667 RETURN(-EOPNOTSUPP);
5669 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5671 rc = cl_falloc(filp, inode, mode, offset, len);
5673 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5674 * used by Lustre in several places. Retuning it here would
5675 * confuse applications that explicity test for EOPNOTSUPP
5676 * (95) and fall back to ftruncate().
5678 if (rc == -ENOTSUPP)
5684 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5685 __u64 start, __u64 len)
5689 struct fiemap *fiemap;
5690 unsigned int extent_count = fieinfo->fi_extents_max;
5692 num_bytes = sizeof(*fiemap) + (extent_count *
5693 sizeof(struct fiemap_extent));
5694 OBD_ALLOC_LARGE(fiemap, num_bytes);
5699 fiemap->fm_flags = fieinfo->fi_flags;
5700 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5701 fiemap->fm_start = start;
5702 fiemap->fm_length = len;
5703 if (extent_count > 0 &&
5704 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5705 sizeof(struct fiemap_extent)) != 0)
5706 GOTO(out, rc = -EFAULT);
5708 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5710 if (IS_ENCRYPTED(inode)) {
5713 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5714 fiemap->fm_extents[i].fe_flags |=
5715 FIEMAP_EXTENT_DATA_ENCRYPTED |
5716 FIEMAP_EXTENT_ENCODED;
5719 fieinfo->fi_flags = fiemap->fm_flags;
5720 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5721 if (extent_count > 0 &&
5722 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5723 fiemap->fm_mapped_extents *
5724 sizeof(struct fiemap_extent)) != 0)
5725 GOTO(out, rc = -EFAULT);
5727 OBD_FREE_LARGE(fiemap, num_bytes);
5731 int ll_inode_permission(struct user_namespace *mnt_userns, struct inode *inode,
5735 struct ll_sb_info *sbi;
5736 struct root_squash_info *squash;
5737 struct cred *cred = NULL;
5738 const struct cred *old_cred = NULL;
5739 bool squash_id = false;
5740 ktime_t kstart = ktime_get();
5744 if (mask & MAY_NOT_BLOCK)
5748 * as root inode are NOT getting validated in lookup operation,
5749 * need to revalidate PERM before permission check.
5751 if (is_root_inode(inode)) {
5752 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_GETATTR);
5757 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5758 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5760 /* squash fsuid/fsgid if needed */
5761 sbi = ll_i2sbi(inode);
5762 squash = &sbi->ll_squash;
5763 if (unlikely(squash->rsi_uid != 0 &&
5764 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5765 !test_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags))) {
5769 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5770 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5771 squash->rsi_uid, squash->rsi_gid);
5773 /* update current process's credentials
5774 * and FS capability */
5775 cred = prepare_creds();
5779 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5780 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5781 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5782 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5784 old_cred = override_creds(cred);
5787 rc = generic_permission(mnt_userns, inode, mask);
5788 /* restore current process's credentials and FS capability */
5790 revert_creds(old_cred);
5795 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5796 ktime_us_delta(ktime_get(), kstart));
5801 /* -o localflock - only provides locally consistent flock locks */
5802 static const struct file_operations ll_file_operations = {
5803 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5804 # ifdef HAVE_SYNC_READ_WRITE
5805 .read = new_sync_read,
5806 .write = new_sync_write,
5808 .read_iter = ll_file_read_iter,
5809 .write_iter = ll_file_write_iter,
5810 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5811 .read = ll_file_read,
5812 .aio_read = ll_file_aio_read,
5813 .write = ll_file_write,
5814 .aio_write = ll_file_aio_write,
5815 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5816 .unlocked_ioctl = ll_file_ioctl,
5817 .open = ll_file_open,
5818 .release = ll_file_release,
5819 .mmap = ll_file_mmap,
5820 .llseek = ll_file_seek,
5821 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5822 .splice_read = generic_file_splice_read,
5824 .splice_read = pcc_file_splice_read,
5826 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5827 .splice_write = iter_file_splice_write,
5831 .fallocate = ll_fallocate,
5834 static const struct file_operations ll_file_operations_flock = {
5835 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5836 # ifdef HAVE_SYNC_READ_WRITE
5837 .read = new_sync_read,
5838 .write = new_sync_write,
5839 # endif /* HAVE_SYNC_READ_WRITE */
5840 .read_iter = ll_file_read_iter,
5841 .write_iter = ll_file_write_iter,
5842 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5843 .read = ll_file_read,
5844 .aio_read = ll_file_aio_read,
5845 .write = ll_file_write,
5846 .aio_write = ll_file_aio_write,
5847 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5848 .unlocked_ioctl = ll_file_ioctl,
5849 .open = ll_file_open,
5850 .release = ll_file_release,
5851 .mmap = ll_file_mmap,
5852 .llseek = ll_file_seek,
5853 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5854 .splice_read = generic_file_splice_read,
5856 .splice_read = pcc_file_splice_read,
5858 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5859 .splice_write = iter_file_splice_write,
5863 .flock = ll_file_flock,
5864 .lock = ll_file_flock,
5865 .fallocate = ll_fallocate,
5868 /* These are for -o noflock - to return ENOSYS on flock calls */
5869 static const struct file_operations ll_file_operations_noflock = {
5870 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5871 # ifdef HAVE_SYNC_READ_WRITE
5872 .read = new_sync_read,
5873 .write = new_sync_write,
5874 # endif /* HAVE_SYNC_READ_WRITE */
5875 .read_iter = ll_file_read_iter,
5876 .write_iter = ll_file_write_iter,
5877 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5878 .read = ll_file_read,
5879 .aio_read = ll_file_aio_read,
5880 .write = ll_file_write,
5881 .aio_write = ll_file_aio_write,
5882 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5883 .unlocked_ioctl = ll_file_ioctl,
5884 .open = ll_file_open,
5885 .release = ll_file_release,
5886 .mmap = ll_file_mmap,
5887 .llseek = ll_file_seek,
5888 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5889 .splice_read = generic_file_splice_read,
5891 .splice_read = pcc_file_splice_read,
5893 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5894 .splice_write = iter_file_splice_write,
5898 .flock = ll_file_noflock,
5899 .lock = ll_file_noflock,
5900 .fallocate = ll_fallocate,
5903 const struct inode_operations ll_file_inode_operations = {
5904 .setattr = ll_setattr,
5905 .getattr = ll_getattr,
5906 .permission = ll_inode_permission,
5907 #ifdef HAVE_IOP_XATTR
5908 .setxattr = ll_setxattr,
5909 .getxattr = ll_getxattr,
5910 .removexattr = ll_removexattr,
5912 .listxattr = ll_listxattr,
5913 .fiemap = ll_fiemap,
5914 .get_acl = ll_get_acl,
5915 #ifdef HAVE_IOP_SET_ACL
5916 .set_acl = ll_set_acl,
5920 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
5922 const struct file_operations *fops = &ll_file_operations_noflock;
5924 if (test_bit(LL_SBI_FLOCK, sbi->ll_flags))
5925 fops = &ll_file_operations_flock;
5926 else if (test_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags))
5927 fops = &ll_file_operations;
5932 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5934 struct ll_inode_info *lli = ll_i2info(inode);
5935 struct cl_object *obj = lli->lli_clob;
5944 env = cl_env_get(&refcheck);
5946 RETURN(PTR_ERR(env));
5948 rc = cl_conf_set(env, lli->lli_clob, conf);
5952 if (conf->coc_opc == OBJECT_CONF_SET) {
5953 struct ldlm_lock *lock = conf->coc_lock;
5954 struct cl_layout cl = {
5958 LASSERT(lock != NULL);
5959 LASSERT(ldlm_has_layout(lock));
5961 /* it can only be allowed to match after layout is
5962 * applied to inode otherwise false layout would be
5963 * seen. Applying layout shoud happen before dropping
5964 * the intent lock. */
5965 ldlm_lock_allow_match(lock);
5967 rc = cl_object_layout_get(env, obj, &cl);
5972 DFID": layout version change: %u -> %u\n",
5973 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5975 ll_layout_version_set(lli, cl.cl_layout_gen);
5979 cl_env_put(env, &refcheck);
5981 RETURN(rc < 0 ? rc : 0);
5984 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5985 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5988 struct ll_sb_info *sbi = ll_i2sbi(inode);
5989 struct ptlrpc_request *req;
5996 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5997 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5998 lock->l_lvb_data, lock->l_lvb_len);
6000 if (lock->l_lvb_data != NULL)
6003 /* if layout lock was granted right away, the layout is returned
6004 * within DLM_LVB of dlm reply; otherwise if the lock was ever
6005 * blocked and then granted via completion ast, we have to fetch
6006 * layout here. Please note that we can't use the LVB buffer in
6007 * completion AST because it doesn't have a large enough buffer */
6008 rc = ll_get_default_mdsize(sbi, &lmmsize);
6012 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
6013 XATTR_NAME_LOV, lmmsize, &req);
6016 GOTO(out, rc = 0); /* empty layout */
6023 if (lmmsize == 0) /* empty layout */
6026 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
6028 GOTO(out, rc = -EFAULT);
6030 OBD_ALLOC_LARGE(lvbdata, lmmsize);
6031 if (lvbdata == NULL)
6032 GOTO(out, rc = -ENOMEM);
6034 memcpy(lvbdata, lmm, lmmsize);
6035 lock_res_and_lock(lock);
6036 if (unlikely(lock->l_lvb_data == NULL)) {
6037 lock->l_lvb_type = LVB_T_LAYOUT;
6038 lock->l_lvb_data = lvbdata;
6039 lock->l_lvb_len = lmmsize;
6042 unlock_res_and_lock(lock);
6045 OBD_FREE_LARGE(lvbdata, lmmsize);
6050 ptlrpc_req_finished(req);
6055 * Apply the layout to the inode. Layout lock is held and will be released
6058 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
6059 struct inode *inode)
6061 struct ll_inode_info *lli = ll_i2info(inode);
6062 struct ll_sb_info *sbi = ll_i2sbi(inode);
6063 struct ldlm_lock *lock;
6064 struct cl_object_conf conf;
6067 bool wait_layout = false;
6070 LASSERT(lustre_handle_is_used(lockh));
6072 lock = ldlm_handle2lock(lockh);
6073 LASSERT(lock != NULL);
6075 if (!ldlm_has_layout(lock))
6076 GOTO(out, rc = -EAGAIN);
6078 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
6079 PFID(&lli->lli_fid), inode);
6081 /* in case this is a caching lock and reinstate with new inode */
6082 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
6084 lock_res_and_lock(lock);
6085 lvb_ready = ldlm_is_lvb_ready(lock);
6086 unlock_res_and_lock(lock);
6088 /* checking lvb_ready is racy but this is okay. The worst case is
6089 * that multi processes may configure the file on the same time. */
6093 rc = ll_layout_fetch(inode, lock);
6097 /* for layout lock, lmm is stored in lock's lvb.
6098 * lvb_data is immutable if the lock is held so it's safe to access it
6101 * set layout to file. Unlikely this will fail as old layout was
6102 * surely eliminated */
6103 memset(&conf, 0, sizeof conf);
6104 conf.coc_opc = OBJECT_CONF_SET;
6105 conf.coc_inode = inode;
6106 conf.coc_lock = lock;
6107 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
6108 conf.u.coc_layout.lb_len = lock->l_lvb_len;
6109 rc = ll_layout_conf(inode, &conf);
6111 /* refresh layout failed, need to wait */
6112 wait_layout = rc == -EBUSY;
6115 LDLM_LOCK_PUT(lock);
6116 ldlm_lock_decref(lockh, mode);
6118 /* wait for IO to complete if it's still being used. */
6120 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
6121 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6123 memset(&conf, 0, sizeof conf);
6124 conf.coc_opc = OBJECT_CONF_WAIT;
6125 conf.coc_inode = inode;
6126 rc = ll_layout_conf(inode, &conf);
6130 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
6131 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
6137 * Issue layout intent RPC to MDS.
6138 * \param inode [in] file inode
6139 * \param intent [in] layout intent
6141 * \retval 0 on success
6142 * \retval < 0 error code
6144 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
6146 struct ll_inode_info *lli = ll_i2info(inode);
6147 struct ll_sb_info *sbi = ll_i2sbi(inode);
6148 struct md_op_data *op_data;
6149 struct lookup_intent it;
6150 struct ptlrpc_request *req;
6154 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
6155 0, 0, LUSTRE_OPC_ANY, NULL);
6156 if (IS_ERR(op_data))
6157 RETURN(PTR_ERR(op_data));
6159 op_data->op_data = intent;
6160 op_data->op_data_size = sizeof(*intent);
6162 memset(&it, 0, sizeof(it));
6163 it.it_op = IT_LAYOUT;
6164 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
6165 intent->li_opc == LAYOUT_INTENT_TRUNC)
6166 it.it_flags = FMODE_WRITE;
6168 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
6169 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6171 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
6172 &ll_md_blocking_ast, 0);
6173 if (it.it_request != NULL)
6174 ptlrpc_req_finished(it.it_request);
6175 it.it_request = NULL;
6177 ll_finish_md_op_data(op_data);
6179 /* set lock data in case this is a new lock */
6181 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
6183 ll_intent_drop_lock(&it);
6189 * This function checks if there exists a LAYOUT lock on the client side,
6190 * or enqueues it if it doesn't have one in cache.
6192 * This function will not hold layout lock so it may be revoked any time after
6193 * this function returns. Any operations depend on layout should be redone
6196 * This function should be called before lov_io_init() to get an uptodate
6197 * layout version, the caller should save the version number and after IO
6198 * is finished, this function should be called again to verify that layout
6199 * is not changed during IO time.
6201 int ll_layout_refresh(struct inode *inode, __u32 *gen)
6203 struct ll_inode_info *lli = ll_i2info(inode);
6204 struct ll_sb_info *sbi = ll_i2sbi(inode);
6205 struct lustre_handle lockh;
6206 struct layout_intent intent = {
6207 .li_opc = LAYOUT_INTENT_ACCESS,
6209 enum ldlm_mode mode;
6213 *gen = ll_layout_version_get(lli);
6214 if (!test_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags) ||
6215 *gen != CL_LAYOUT_GEN_NONE)
6219 LASSERT(fid_is_sane(ll_inode2fid(inode)));
6220 LASSERT(S_ISREG(inode->i_mode));
6222 /* take layout lock mutex to enqueue layout lock exclusively. */
6223 mutex_lock(&lli->lli_layout_mutex);
6226 /* mostly layout lock is caching on the local side, so try to
6227 * match it before grabbing layout lock mutex. */
6228 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
6229 LCK_CR | LCK_CW | LCK_PR |
6231 if (mode != 0) { /* hit cached lock */
6232 rc = ll_layout_lock_set(&lockh, mode, inode);
6238 rc = ll_layout_intent(inode, &intent);
6244 *gen = ll_layout_version_get(lli);
6245 mutex_unlock(&lli->lli_layout_mutex);
6251 * Issue layout intent RPC indicating where in a file an IO is about to write.
6253 * \param[in] inode file inode.
6254 * \param[in] ext write range with start offset of fille in bytes where
6255 * an IO is about to write, and exclusive end offset in
6258 * \retval 0 on success
6259 * \retval < 0 error code
6261 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
6262 struct lu_extent *ext)
6264 struct layout_intent intent = {
6266 .li_extent.e_start = ext->e_start,
6267 .li_extent.e_end = ext->e_end,
6272 rc = ll_layout_intent(inode, &intent);
6278 * This function send a restore request to the MDT
6280 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6282 struct hsm_user_request *hur;
6286 len = sizeof(struct hsm_user_request) +
6287 sizeof(struct hsm_user_item);
6288 OBD_ALLOC(hur, len);
6292 hur->hur_request.hr_action = HUA_RESTORE;
6293 hur->hur_request.hr_archive_id = 0;
6294 hur->hur_request.hr_flags = 0;
6295 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6296 sizeof(hur->hur_user_item[0].hui_fid));
6297 hur->hur_user_item[0].hui_extent.offset = offset;
6298 hur->hur_user_item[0].hui_extent.length = length;
6299 hur->hur_request.hr_itemcount = 1;
6300 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,