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 static int ll_merge_attr_nolock(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 /* Merge timestamps the most recently obtained from MDS with
1437 * timestamps obtained from OSTs.
1439 * Do not overwrite atime of inode because it may be refreshed
1440 * by file_accessed() function. If the read was served by cache
1441 * data, there is no RPC to be sent so that atime may not be
1442 * transferred to OSTs at all. MDT only updates atime at close time
1443 * if it's at least 'mdd.*.atime_diff' older.
1444 * All in all, the atime in Lustre does not strictly comply with
1445 * POSIX. Solving this problem needs to send an RPC to MDT for each
1446 * read, this will hurt performance.
1448 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1449 inode->i_atime.tv_sec < lli->lli_atime)
1450 inode->i_atime.tv_sec = lli->lli_atime;
1452 inode->i_mtime.tv_sec = lli->lli_mtime;
1453 inode->i_ctime.tv_sec = lli->lli_ctime;
1455 mtime = inode->i_mtime.tv_sec;
1456 atime = inode->i_atime.tv_sec;
1457 ctime = inode->i_ctime.tv_sec;
1459 cl_object_attr_lock(obj);
1460 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1463 rc = cl_object_attr_get(env, obj, attr);
1464 cl_object_attr_unlock(obj);
1467 GOTO(out, rc = (rc == -ENODATA ? 0 : rc));
1469 if (atime < attr->cat_atime)
1470 atime = attr->cat_atime;
1472 if (ctime < attr->cat_ctime)
1473 ctime = attr->cat_ctime;
1475 if (mtime < attr->cat_mtime)
1476 mtime = attr->cat_mtime;
1478 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu i_blocks %llu\n",
1479 PFID(&lli->lli_fid), attr->cat_size, attr->cat_blocks);
1481 if (llcrypt_require_key(inode) == -ENOKEY) {
1482 /* Without the key, round up encrypted file size to next
1483 * LUSTRE_ENCRYPTION_UNIT_SIZE. Clear text size is put in
1484 * lli_lazysize for proper file size setting at close time.
1486 lli->lli_attr_valid |= OBD_MD_FLLAZYSIZE;
1487 lli->lli_lazysize = attr->cat_size;
1488 attr->cat_size = round_up(attr->cat_size,
1489 LUSTRE_ENCRYPTION_UNIT_SIZE);
1491 i_size_write(inode, attr->cat_size);
1492 inode->i_blocks = attr->cat_blocks;
1494 inode->i_mtime.tv_sec = mtime;
1495 inode->i_atime.tv_sec = atime;
1496 inode->i_ctime.tv_sec = ctime;
1503 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1507 ll_inode_size_lock(inode);
1508 rc = ll_merge_attr_nolock(env, inode);
1509 ll_inode_size_unlock(inode);
1514 /* Use to update size and blocks on inode for LSOM if there is no contention */
1515 int ll_merge_attr_try(const struct lu_env *env, struct inode *inode)
1519 if (ll_inode_size_trylock(inode)) {
1520 rc = ll_merge_attr_nolock(env, inode);
1521 ll_inode_size_unlock(inode);
1528 * Set designated mirror for I/O.
1530 * So far only read, write, and truncated can support to issue I/O to
1531 * designated mirror.
1533 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1535 struct ll_file_data *fd = file->private_data;
1537 /* clear layout version for generic(non-resync) I/O in case it carries
1538 * stale layout version due to I/O restart */
1539 io->ci_layout_version = 0;
1541 /* FLR: disable non-delay for designated mirror I/O because obviously
1542 * only one mirror is available */
1543 if (fd->fd_designated_mirror > 0) {
1545 io->ci_designated_mirror = fd->fd_designated_mirror;
1546 io->ci_layout_version = fd->fd_layout_version;
1549 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1550 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1554 * This is relatime_need_update() from Linux 5.17, which is not exported.
1556 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1557 struct timespec64 now)
1560 if (!(mnt->mnt_flags & MNT_RELATIME))
1563 * Is mtime younger than atime? If yes, update atime:
1565 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1568 * Is ctime younger than atime? If yes, update atime:
1570 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1574 * Is the previous atime value older than a day? If yes,
1577 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1580 * Good, we can skip the atime update:
1586 * Very similar to kernel function: !__atime_needs_update()
1588 static bool file_is_noatime(const struct file *file)
1590 struct vfsmount *mnt = file->f_path.mnt;
1591 struct inode *inode = file_inode((struct file *)file);
1592 struct timespec64 now;
1594 if (file->f_flags & O_NOATIME)
1597 if (inode->i_flags & S_NOATIME)
1600 if (IS_NOATIME(inode))
1603 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1606 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1609 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1612 now = current_time(inode);
1614 if (!relatime_need_update(mnt, inode, now))
1620 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1621 struct vvp_io_args *args)
1623 struct inode *inode = file_inode(file);
1624 struct ll_file_data *fd = file->private_data;
1626 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1627 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1629 if (iot == CIT_WRITE) {
1630 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1631 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1632 file->f_flags & O_DIRECT ||
1634 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1635 io->u.ci_wr.wr_sync |= !!(args &&
1636 (args->u.normal.via_iocb->ki_flags &
1642 io->ci_iocb_nowait = !!(args &&
1643 (args->u.normal.via_iocb->ki_flags &
1647 io->ci_obj = ll_i2info(inode)->lli_clob;
1648 io->ci_lockreq = CILR_MAYBE;
1649 if (ll_file_nolock(file)) {
1650 io->ci_lockreq = CILR_NEVER;
1651 io->ci_no_srvlock = 1;
1652 } else if (file->f_flags & O_APPEND) {
1653 io->ci_lockreq = CILR_MANDATORY;
1655 io->ci_noatime = file_is_noatime(file);
1656 io->ci_async_readahead = false;
1658 /* FLR: only use non-delay I/O for read as there is only one
1659 * avaliable mirror for write. */
1660 io->ci_ndelay = !(iot == CIT_WRITE);
1662 ll_io_set_mirror(io, file);
1665 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1668 struct ll_inode_info *lli = ll_i2info(inode);
1669 struct ll_sb_info *sbi = ll_i2sbi(inode);
1670 enum obd_heat_type sample_type;
1671 enum obd_heat_type iobyte_type;
1672 __u64 now = ktime_get_real_seconds();
1674 if (!ll_sbi_has_file_heat(sbi) ||
1675 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1678 if (iot == CIT_READ) {
1679 sample_type = OBD_HEAT_READSAMPLE;
1680 iobyte_type = OBD_HEAT_READBYTE;
1681 } else if (iot == CIT_WRITE) {
1682 sample_type = OBD_HEAT_WRITESAMPLE;
1683 iobyte_type = OBD_HEAT_WRITEBYTE;
1688 spin_lock(&lli->lli_heat_lock);
1689 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1690 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1691 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1692 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1693 spin_unlock(&lli->lli_heat_lock);
1697 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1698 struct file *file, enum cl_io_type iot,
1699 loff_t *ppos, size_t count)
1701 struct vvp_io *vio = vvp_env_io(env);
1702 struct inode *inode = file_inode(file);
1703 struct ll_inode_info *lli = ll_i2info(inode);
1704 struct ll_sb_info *sbi = ll_i2sbi(inode);
1705 struct ll_file_data *fd = file->private_data;
1706 struct range_lock range;
1707 bool range_locked = false;
1713 unsigned int retried = 0, dio_lock = 0;
1714 bool is_aio = false;
1715 bool is_parallel_dio = false;
1716 struct cl_dio_aio *ci_dio_aio = NULL;
1718 bool partial_io = false;
1719 size_t max_io_pages, max_cached_pages;
1723 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1724 file_dentry(file)->d_name.name,
1725 iot == CIT_READ ? "read" : "write", *ppos, count);
1727 max_io_pages = PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT;
1728 max_cached_pages = sbi->ll_cache->ccc_lru_max;
1729 if (max_io_pages > (max_cached_pages >> 2))
1730 max_io_pages = max_cached_pages >> 2;
1732 io = vvp_env_thread_io(env);
1733 if (file->f_flags & O_DIRECT) {
1734 if (file->f_flags & O_APPEND)
1736 if (!is_sync_kiocb(args->u.normal.via_iocb))
1739 /* the kernel does not support AIO on pipes, and parallel DIO
1740 * uses part of the AIO path, so we must not do parallel dio
1743 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1746 if (!ll_sbi_has_parallel_dio(sbi))
1747 is_parallel_dio = false;
1749 ci_dio_aio = cl_dio_aio_alloc(args->u.normal.via_iocb,
1750 ll_i2info(inode)->lli_clob, is_aio);
1752 GOTO(out, rc = -ENOMEM);
1757 * IO block size need be aware of cached page limit, otherwise
1758 * if we have small max_cached_mb but large block IO issued, io
1759 * could not be finished and blocked whole client.
1761 if (file->f_flags & O_DIRECT)
1764 per_bytes = min(max_io_pages << PAGE_SHIFT, count);
1765 partial_io = per_bytes < count;
1766 io = vvp_env_thread_io(env);
1767 ll_io_init(io, file, iot, args);
1768 io->ci_dio_aio = ci_dio_aio;
1769 io->ci_dio_lock = dio_lock;
1770 io->ci_ndelay_tried = retried;
1771 io->ci_parallel_dio = is_parallel_dio;
1773 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1774 if (file->f_flags & O_APPEND)
1775 range_lock_init(&range, 0, LUSTRE_EOF);
1777 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1779 vio->vui_fd = file->private_data;
1780 vio->vui_iter = args->u.normal.via_iter;
1781 vio->vui_iocb = args->u.normal.via_iocb;
1782 /* Direct IO reads must also take range lock,
1783 * or multiple reads will try to work on the same pages
1784 * See LU-6227 for details.
1786 if (((iot == CIT_WRITE) ||
1787 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1788 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1789 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1791 rc = range_lock(&lli->lli_write_tree, &range);
1795 range_locked = true;
1798 ll_cl_add(inode, env, io, LCC_RW);
1799 rc = cl_io_loop(env, io);
1800 ll_cl_remove(inode, env);
1802 /* cl_io_rw_init() handled IO */
1806 if (io->ci_dio_aio && !is_aio) {
1807 struct cl_sync_io *anchor = &io->ci_dio_aio->cda_sync;
1809 /* for dio, EIOCBQUEUED is an implementation detail,
1810 * and we don't return it to userspace
1812 if (rc == -EIOCBQUEUED)
1815 /* N/B: parallel DIO may be disabled during i/o submission;
1816 * if that occurs, I/O shifts to sync, so it's all resolved
1817 * before we get here, and this wait call completes
1820 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1826 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1828 range_unlock(&lli->lli_write_tree, &range);
1829 range_locked = false;
1833 * In order to move forward AIO, ci_nob was increased,
1834 * but that doesn't mean io have been finished, it just
1835 * means io have been submited, we will always return
1836 * EIOCBQUEUED to the caller, So we could only return
1837 * number of bytes in non-AIO case.
1839 if (io->ci_nob > 0) {
1842 result += io->ci_nob;
1843 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1848 count -= io->ci_nob;
1850 /* prepare IO restart */
1852 args->u.normal.via_iter = vio->vui_iter;
1856 * Reexpand iov count because it was zero
1859 iov_iter_reexpand(vio->vui_iter, count);
1860 if (per_bytes == io->ci_nob)
1861 io->ci_need_restart = 1;
1865 cl_io_fini(env, io);
1868 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1869 file->f_path.dentry->d_name.name,
1870 iot, rc, result, io->ci_need_restart);
1872 if ((!rc || rc == -ENODATA || rc == -ENOLCK || rc == -EIOCBQUEUED) &&
1873 count > 0 && io->ci_need_restart && retries-- > 0) {
1875 "%s: restart %s from ppos=%lld count=%zu retries=%u ret=%zd: rc = %d\n",
1876 file_dentry(file)->d_name.name,
1877 iot == CIT_READ ? "read" : "write",
1878 *ppos, count, retries, result, rc);
1879 /* preserve the tried count for FLR */
1880 retried = io->ci_ndelay_tried;
1881 dio_lock = io->ci_dio_lock;
1885 if (io->ci_dio_aio) {
1887 * VFS will call aio_complete() if no -EIOCBQUEUED
1888 * is returned for AIO, so we can not call aio_complete()
1891 * NB: This is safe because the atomic_dec_and_lock in
1892 * cl_sync_io_init has implicit memory barriers, so this will
1893 * be seen by whichever thread completes the DIO/AIO, even if
1896 if (rc != -EIOCBQUEUED)
1897 io->ci_dio_aio->cda_no_aio_complete = 1;
1899 * Drop one extra reference so that end_io() could be
1900 * called for this IO context, we could call it after
1901 * we make sure all AIO requests have been proceed.
1903 cl_sync_io_note(env, &io->ci_dio_aio->cda_sync,
1904 rc == -EIOCBQUEUED ? 0 : rc);
1906 LASSERT(io->ci_dio_aio->cda_creator_free);
1907 cl_dio_aio_free(env, io->ci_dio_aio);
1908 io->ci_dio_aio = NULL;
1912 if (iot == CIT_READ) {
1914 ll_stats_ops_tally(ll_i2sbi(inode),
1915 LPROC_LL_READ_BYTES, result);
1916 } else if (iot == CIT_WRITE) {
1918 ll_stats_ops_tally(ll_i2sbi(inode),
1919 LPROC_LL_WRITE_BYTES, result);
1920 fd->fd_write_failed = false;
1921 } else if (result == 0 && rc == 0) {
1924 fd->fd_write_failed = true;
1926 fd->fd_write_failed = false;
1927 } else if (rc != -ERESTARTSYS) {
1928 fd->fd_write_failed = true;
1932 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1934 ll_heat_add(inode, iot, result);
1936 RETURN(result > 0 ? result : rc);
1940 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1941 * especially for small I/O.
1943 * To serve a read request, CLIO has to create and initialize a cl_io and
1944 * then request DLM lock. This has turned out to have siginificant overhead
1945 * and affects the performance of small I/O dramatically.
1947 * It's not necessary to create a cl_io for each I/O. Under the help of read
1948 * ahead, most of the pages being read are already in memory cache and we can
1949 * read those pages directly because if the pages exist, the corresponding DLM
1950 * lock must exist so that page content must be valid.
1952 * In fast read implementation, the llite speculatively finds and reads pages
1953 * in memory cache. There are three scenarios for fast read:
1954 * - If the page exists and is uptodate, kernel VM will provide the data and
1955 * CLIO won't be intervened;
1956 * - If the page was brought into memory by read ahead, it will be exported
1957 * and read ahead parameters will be updated;
1958 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1959 * it will go back and invoke normal read, i.e., a cl_io will be created
1960 * and DLM lock will be requested.
1962 * POSIX compliance: posix standard states that read is intended to be atomic.
1963 * Lustre read implementation is in line with Linux kernel read implementation
1964 * and neither of them complies with POSIX standard in this matter. Fast read
1965 * doesn't make the situation worse on single node but it may interleave write
1966 * results from multiple nodes due to short read handling in ll_file_aio_read().
1968 * \param env - lu_env
1969 * \param iocb - kiocb from kernel
1970 * \param iter - user space buffers where the data will be copied
1972 * \retval - number of bytes have been read, or error code if error occurred.
1975 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1979 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1982 /* NB: we can't do direct IO for fast read because it will need a lock
1983 * to make IO engine happy. */
1984 if (iocb->ki_filp->f_flags & O_DIRECT)
1987 result = generic_file_read_iter(iocb, iter);
1989 /* If the first page is not in cache, generic_file_aio_read() will be
1990 * returned with -ENODATA.
1991 * See corresponding code in ll_readpage(). */
1992 if (result == -ENODATA)
1996 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1997 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1998 LPROC_LL_READ_BYTES, result);
2005 * Confine read iter lest read beyond the EOF
2007 * \param iocb [in] kernel iocb
2008 * \param to [in] reader iov_iter
2010 * \retval <0 failure
2012 * \retval >0 @iocb->ki_pos has passed the EOF
2014 static int file_read_confine_iter(struct lu_env *env, struct kiocb *iocb,
2015 struct iov_iter *to)
2017 struct cl_attr *attr = vvp_env_thread_attr(env);
2018 struct file *file = iocb->ki_filp;
2019 struct inode *inode = file_inode(file);
2020 struct ll_inode_info *lli = ll_i2info(inode);
2021 loff_t read_end = iocb->ki_pos + iov_iter_count(to);
2026 cl_object_attr_lock(lli->lli_clob);
2027 rc = cl_object_attr_get(env, lli->lli_clob, attr);
2028 cl_object_attr_unlock(lli->lli_clob);
2032 kms = attr->cat_kms;
2033 /* if read beyond end-of-file, adjust read count */
2034 if (kms > 0 && (iocb->ki_pos >= kms || read_end > kms)) {
2035 rc = ll_glimpse_size(inode);
2039 size = i_size_read(inode);
2040 if (iocb->ki_pos >= size || read_end > size) {
2042 "%s: read [%llu, %llu] over eof, kms %llu, file_size %llu.\n",
2043 file_dentry(file)->d_name.name,
2044 iocb->ki_pos, read_end, kms, size);
2046 if (iocb->ki_pos >= size)
2049 if (read_end > size)
2050 iov_iter_truncate(to, size - iocb->ki_pos);
2058 * Read from a file (through the page cache).
2060 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2063 struct vvp_io_args *args;
2064 struct file *file = iocb->ki_filp;
2068 ktime_t kstart = ktime_get();
2070 bool stale_data = false;
2074 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2075 file_dentry(file)->d_name.name,
2076 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2077 iov_iter_count(to));
2079 if (!iov_iter_count(to))
2082 env = cl_env_get(&refcheck);
2084 RETURN(PTR_ERR(env));
2086 result = file_read_confine_iter(env, iocb, to);
2089 else if (result > 0)
2093 * Currently when PCC read failed, we do not fall back to the
2094 * normal read path, just return the error.
2095 * The resaon is that: for RW-PCC, the file data may be modified
2096 * in the PCC and inconsistent with the data on OSTs (or file
2097 * data has been removed from the Lustre file system), at this
2098 * time, fallback to the normal read path may read the wrong
2100 * TODO: for RO-PCC (readonly PCC), fall back to normal read
2101 * path: read data from data copy on OSTs.
2103 result = pcc_file_read_iter(iocb, to, &cached);
2107 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
2109 result = ll_do_fast_read(iocb, to);
2110 if (result < 0 || iov_iter_count(to) == 0)
2113 args = ll_env_args(env);
2114 args->u.normal.via_iter = to;
2115 args->u.normal.via_iocb = iocb;
2117 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
2118 &iocb->ki_pos, iov_iter_count(to));
2121 else if (result == 0)
2125 cl_env_put(env, &refcheck);
2127 if (stale_data && result > 0) {
2129 * we've reached EOF before the read, the data read are cached
2132 iov_iter_truncate(to, 0);
2137 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2138 file->private_data, iocb->ki_pos, result,
2140 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
2141 ktime_us_delta(ktime_get(), kstart));
2145 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2146 file_dentry(file)->d_name.name,
2147 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2148 iov_iter_count(to));
2154 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
2155 * If a page is already in the page cache and dirty (and some other things -
2156 * See ll_tiny_write_begin for the instantiation of these rules), then we can
2157 * write to it without doing a full I/O, because Lustre already knows about it
2158 * and will write it out. This saves a lot of processing time.
2160 * All writes here are within one page, so exclusion is handled by the page
2161 * lock on the vm page. We do not do tiny writes for writes which touch
2162 * multiple pages because it's very unlikely multiple sequential pages are
2163 * are already dirty.
2165 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
2166 * and are unlikely to be to already dirty pages.
2168 * Attribute updates are important here, we do them in ll_tiny_write_end.
2170 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
2172 ssize_t count = iov_iter_count(iter);
2173 struct file *file = iocb->ki_filp;
2174 struct inode *inode = file_inode(file);
2175 bool lock_inode = !IS_NOSEC(inode);
2180 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2181 * of function for why.
2183 if (count >= PAGE_SIZE ||
2184 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2187 if (unlikely(lock_inode))
2189 result = __generic_file_write_iter(iocb, iter);
2191 if (unlikely(lock_inode))
2192 inode_unlock(inode);
2194 /* If the page is not already dirty, ll_tiny_write_begin returns
2195 * -ENODATA. We continue on to normal write.
2197 if (result == -ENODATA)
2201 ll_heat_add(inode, CIT_WRITE, result);
2202 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2205 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2211 * Write to a file (through the page cache).
2213 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2215 struct vvp_io_args *args;
2217 ssize_t rc_tiny = 0, rc_normal;
2218 struct file *file = iocb->ki_filp;
2221 ktime_t kstart = ktime_get();
2226 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2227 file_dentry(file)->d_name.name,
2228 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2229 iov_iter_count(from));
2231 if (!iov_iter_count(from))
2232 GOTO(out, rc_normal = 0);
2235 * When PCC write failed, we usually do not fall back to the normal
2236 * write path, just return the error. But there is a special case when
2237 * returned error code is -ENOSPC due to running out of space on PCC HSM
2238 * bakcend. At this time, it will fall back to normal I/O path and
2239 * retry the I/O. As the file is in HSM released state, it will restore
2240 * the file data to OSTs first and redo the write again. And the
2241 * restore process will revoke the layout lock and detach the file
2242 * from PCC cache automatically.
2244 result = pcc_file_write_iter(iocb, from, &cached);
2245 if (cached && result != -ENOSPC && result != -EDQUOT)
2246 GOTO(out, rc_normal = result);
2248 /* NB: we can't do direct IO for tiny writes because they use the page
2249 * cache, we can't do sync writes because tiny writes can't flush
2250 * pages, and we can't do append writes because we can't guarantee the
2251 * required DLM locks are held to protect file size.
2253 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2254 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
2255 rc_tiny = ll_do_tiny_write(iocb, from);
2257 /* In case of error, go on and try normal write - Only stop if tiny
2258 * write completed I/O.
2260 if (iov_iter_count(from) == 0)
2261 GOTO(out, rc_normal = rc_tiny);
2263 env = cl_env_get(&refcheck);
2265 RETURN(PTR_ERR(env));
2267 args = ll_env_args(env);
2268 args->u.normal.via_iter = from;
2269 args->u.normal.via_iocb = iocb;
2271 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2272 &iocb->ki_pos, iov_iter_count(from));
2274 /* On success, combine bytes written. */
2275 if (rc_tiny >= 0 && rc_normal > 0)
2276 rc_normal += rc_tiny;
2277 /* On error, only return error from normal write if tiny write did not
2278 * write any bytes. Otherwise return bytes written by tiny write.
2280 else if (rc_tiny > 0)
2281 rc_normal = rc_tiny;
2283 cl_env_put(env, &refcheck);
2285 if (rc_normal > 0) {
2286 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2287 file->private_data, iocb->ki_pos,
2289 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2290 ktime_us_delta(ktime_get(), kstart));
2294 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2295 file_dentry(file)->d_name.name,
2296 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2297 iov_iter_count(from));
2302 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2304 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2306 static int ll_file_get_iov_count(const struct iovec *iov,
2307 unsigned long *nr_segs, size_t *count,
2313 for (seg = 0; seg < *nr_segs; seg++) {
2314 const struct iovec *iv = &iov[seg];
2317 * If any segment has a negative length, or the cumulative
2318 * length ever wraps negative then return -EINVAL.
2321 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2323 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2328 cnt -= iv->iov_len; /* This segment is no good */
2335 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2336 unsigned long nr_segs, loff_t pos)
2343 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2350 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2351 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2352 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2353 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2354 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2356 result = ll_file_read_iter(iocb, &to);
2361 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2364 struct iovec iov = { .iov_base = buf, .iov_len = count };
2373 init_sync_kiocb(&kiocb, file);
2374 kiocb.ki_pos = *ppos;
2375 #ifdef HAVE_KIOCB_KI_LEFT
2376 kiocb.ki_left = count;
2377 #elif defined(HAVE_KI_NBYTES)
2378 kiocb.i_nbytes = count;
2381 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2382 *ppos = kiocb.ki_pos;
2388 * Write to a file (through the page cache).
2391 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2392 unsigned long nr_segs, loff_t pos)
2394 struct iov_iter from;
2399 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2406 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2407 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2408 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2409 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2410 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2412 result = ll_file_write_iter(iocb, &from);
2417 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2418 size_t count, loff_t *ppos)
2420 struct iovec iov = { .iov_base = (void __user *)buf,
2430 init_sync_kiocb(&kiocb, file);
2431 kiocb.ki_pos = *ppos;
2432 #ifdef HAVE_KIOCB_KI_LEFT
2433 kiocb.ki_left = count;
2434 #elif defined(HAVE_KI_NBYTES)
2435 kiocb.ki_nbytes = count;
2438 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2439 *ppos = kiocb.ki_pos;
2443 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2445 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2446 __u64 flags, struct lov_user_md *lum, int lum_size)
2448 struct lookup_intent oit = {
2450 .it_flags = flags | MDS_OPEN_BY_FID,
2455 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2456 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2457 /* this code will only exist for big-endian systems */
2458 lustre_swab_lov_user_md(lum, 0);
2461 ll_inode_size_lock(inode);
2462 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2464 GOTO(out_unlock, rc);
2466 ll_release_openhandle(dentry, &oit);
2469 ll_inode_size_unlock(inode);
2470 ll_intent_release(&oit);
2475 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2476 struct lov_mds_md **lmmp, int *lmm_size,
2477 struct ptlrpc_request **request)
2479 struct ll_sb_info *sbi = ll_i2sbi(inode);
2480 struct mdt_body *body;
2481 struct lov_mds_md *lmm = NULL;
2482 struct ptlrpc_request *req = NULL;
2483 struct md_op_data *op_data;
2488 rc = ll_get_default_mdsize(sbi, &lmmsize);
2492 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2493 strlen(filename), lmmsize,
2494 LUSTRE_OPC_ANY, NULL);
2495 if (IS_ERR(op_data))
2496 RETURN(PTR_ERR(op_data));
2498 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2499 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2500 ll_finish_md_op_data(op_data);
2502 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2507 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2508 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2510 lmmsize = body->mbo_eadatasize;
2512 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2514 GOTO(out, rc = -ENODATA);
2516 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2517 LASSERT(lmm != NULL);
2519 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2520 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2521 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2522 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2523 GOTO(out, rc = -EPROTO);
2526 * This is coming from the MDS, so is probably in
2527 * little endian. We convert it to host endian before
2528 * passing it to userspace.
2530 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2531 int stripe_count = 0;
2533 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2534 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2535 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2536 if (le32_to_cpu(lmm->lmm_pattern) &
2537 LOV_PATTERN_F_RELEASED)
2539 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2541 /* if function called for directory - we should
2542 * avoid swab not existent lsm objects
2544 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2545 S_ISREG(body->mbo_mode))
2546 lustre_swab_lov_user_md_objects(
2547 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2549 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2550 S_ISREG(body->mbo_mode))
2551 lustre_swab_lov_user_md_objects(
2552 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2554 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2555 lustre_swab_lov_comp_md_v1(
2556 (struct lov_comp_md_v1 *)lmm);
2560 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2561 struct lov_comp_md_v1 *comp_v1 = NULL;
2562 struct lov_comp_md_entry_v1 *ent;
2563 struct lov_user_md_v1 *v1;
2567 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2568 /* Dump the striping information */
2569 for (; i < comp_v1->lcm_entry_count; i++) {
2570 ent = &comp_v1->lcm_entries[i];
2571 off = ent->lcme_offset;
2572 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2574 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2575 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2579 * Return valid stripe_count and stripe_size instead of 0 for
2580 * DoM files to avoid divide-by-zero for older userspace that
2581 * calls this ioctl, e.g. lustre ADIO driver.
2583 if (lmm->lmm_stripe_count == 0)
2584 lmm->lmm_stripe_count = 1;
2585 if (lmm->lmm_stripe_size == 0) {
2586 /* Since the first component of the file data is placed
2587 * on the MDT for faster access, the stripe_size of the
2588 * second one is always that applications which are
2591 if (lmm->lmm_pattern == LOV_PATTERN_MDT)
2592 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2594 i = comp_v1->lcm_entry_count > 1 ?
2595 comp_v1->lcm_entry_count - 1 : 0;
2596 ent = &comp_v1->lcm_entries[i];
2597 off = ent->lcme_offset;
2598 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2599 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2604 *lmm_size = lmmsize;
2609 static int ll_lov_setea(struct inode *inode, struct file *file,
2612 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2613 struct lov_user_md *lump;
2614 int lum_size = sizeof(*lump) + sizeof(struct lov_user_ost_data);
2618 if (!capable(CAP_SYS_ADMIN))
2621 OBD_ALLOC_LARGE(lump, lum_size);
2625 if (copy_from_user(lump, arg, lum_size))
2626 GOTO(out_lump, rc = -EFAULT);
2628 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2630 cl_lov_delay_create_clear(&file->f_flags);
2633 OBD_FREE_LARGE(lump, lum_size);
2637 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2644 env = cl_env_get(&refcheck);
2646 RETURN(PTR_ERR(env));
2648 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2649 cl_env_put(env, &refcheck);
2653 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2656 struct lov_user_md __user *lum = arg;
2657 struct lov_user_md *klum;
2659 __u64 flags = FMODE_WRITE;
2662 rc = ll_copy_user_md(lum, &klum);
2667 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2672 rc = put_user(0, &lum->lmm_stripe_count);
2676 rc = ll_layout_refresh(inode, &gen);
2680 rc = ll_file_getstripe(inode, arg, lum_size);
2681 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2682 ll_i2info(inode)->lli_clob) {
2683 struct iattr attr = { 0 };
2685 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2686 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2689 cl_lov_delay_create_clear(&file->f_flags);
2692 OBD_FREE_LARGE(klum, lum_size);
2698 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2700 struct ll_inode_info *lli = ll_i2info(inode);
2701 struct cl_object *obj = lli->lli_clob;
2702 struct ll_file_data *fd = file->private_data;
2703 struct ll_grouplock grouplock;
2708 CWARN("group id for group lock must not be 0\n");
2712 if (ll_file_nolock(file))
2713 RETURN(-EOPNOTSUPP);
2715 if (file->f_flags & O_NONBLOCK) {
2716 if (!mutex_trylock(&lli->lli_group_mutex))
2719 mutex_lock(&lli->lli_group_mutex);
2722 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2723 CWARN("group lock already existed with gid %lu\n",
2724 fd->fd_grouplock.lg_gid);
2725 GOTO(out, rc = -EINVAL);
2727 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2728 if (file->f_flags & O_NONBLOCK)
2729 GOTO(out, rc = -EAGAIN);
2730 mutex_unlock(&lli->lli_group_mutex);
2731 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2732 GOTO(retry, rc = 0);
2734 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2737 * XXX: group lock needs to protect all OST objects while PFL
2738 * can add new OST objects during the IO, so we'd instantiate
2739 * all OST objects before getting its group lock.
2744 struct cl_layout cl = {
2745 .cl_is_composite = false,
2747 struct lu_extent ext = {
2749 .e_end = OBD_OBJECT_EOF,
2752 env = cl_env_get(&refcheck);
2754 GOTO(out, rc = PTR_ERR(env));
2756 rc = cl_object_layout_get(env, obj, &cl);
2757 if (rc >= 0 && cl.cl_is_composite)
2758 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2761 cl_env_put(env, &refcheck);
2766 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2767 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2772 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2773 fd->fd_grouplock = grouplock;
2774 if (lli->lli_group_users == 0)
2775 lli->lli_group_gid = grouplock.lg_gid;
2776 lli->lli_group_users++;
2778 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2780 mutex_unlock(&lli->lli_group_mutex);
2785 static int ll_put_grouplock(struct inode *inode, struct file *file,
2788 struct ll_inode_info *lli = ll_i2info(inode);
2789 struct ll_file_data *fd = file->private_data;
2790 struct ll_grouplock grouplock;
2794 mutex_lock(&lli->lli_group_mutex);
2795 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2796 CWARN("no group lock held\n");
2797 GOTO(out, rc = -EINVAL);
2800 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2802 if (fd->fd_grouplock.lg_gid != arg) {
2803 CWARN("group lock %lu doesn't match current id %lu\n",
2804 arg, fd->fd_grouplock.lg_gid);
2805 GOTO(out, rc = -EINVAL);
2808 grouplock = fd->fd_grouplock;
2809 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2810 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2812 cl_put_grouplock(&grouplock);
2814 lli->lli_group_users--;
2815 if (lli->lli_group_users == 0) {
2816 lli->lli_group_gid = 0;
2817 wake_up_var(&lli->lli_group_users);
2819 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2822 mutex_unlock(&lli->lli_group_mutex);
2828 * Close inode open handle
2830 * \param dentry [in] dentry which contains the inode
2831 * \param it [in,out] intent which contains open info and result
2834 * \retval <0 failure
2836 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2838 struct inode *inode = dentry->d_inode;
2839 struct obd_client_handle *och;
2845 /* Root ? Do nothing. */
2846 if (is_root_inode(inode))
2849 /* No open handle to close? Move away */
2850 if (!it_disposition(it, DISP_OPEN_OPEN))
2853 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2855 OBD_ALLOC(och, sizeof(*och));
2857 GOTO(out, rc = -ENOMEM);
2859 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2863 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2865 /* this one is in place of ll_file_open */
2866 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2867 ptlrpc_req_finished(it->it_request);
2868 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2874 * Get size for inode for which FIEMAP mapping is requested.
2875 * Make the FIEMAP get_info call and returns the result.
2876 * \param fiemap kernel buffer to hold extens
2877 * \param num_bytes kernel buffer size
2879 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2885 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2888 /* Checks for fiemap flags */
2889 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2890 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2894 /* Check for FIEMAP_FLAG_SYNC */
2895 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2896 rc = filemap_fdatawrite(inode->i_mapping);
2901 env = cl_env_get(&refcheck);
2903 RETURN(PTR_ERR(env));
2905 if (i_size_read(inode) == 0) {
2906 rc = ll_glimpse_size(inode);
2911 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLPROJID;
2912 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2913 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2915 /* If filesize is 0, then there would be no objects for mapping */
2916 if (fmkey.lfik_oa.o_size == 0) {
2917 fiemap->fm_mapped_extents = 0;
2921 fmkey.lfik_fiemap = *fiemap;
2923 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2924 &fmkey, fiemap, &num_bytes);
2926 cl_env_put(env, &refcheck);
2930 static int fid2path_for_enc_file(struct inode *parent, char *gfpath,
2933 struct dentry *de = NULL, *de_parent = d_find_any_alias(parent);
2934 struct llcrypt_str lltr = LLTR_INIT(NULL, 0);
2935 struct llcrypt_str de_name;
2936 char *p, *ptr = gfpath;
2937 size_t len = 0, len_orig = 0;
2938 int enckey = -1, nameenc = -1;
2942 while ((p = strsep(&gfpath, "/")) != NULL) {
2950 len_orig = strlen(p);
2952 rc = sscanf(p, "["SFID"]", RFID(&fid));
2954 p = strchr(p, ']') + 1;
2960 if (!IS_ENCRYPTED(parent)) {
2961 if (gfpathlen < len + 1) {
2966 memmove(ptr, p, len);
2970 gfpathlen -= len + 1;
2974 /* From here, we know parent is encrypted */
2977 rc = llcrypt_prepare_readdir(parent);
2978 if (rc && rc != -ENOKEY) {
2985 if (llcrypt_has_encryption_key(parent))
2991 llcrypt_policy_has_filename_enc(parent);
2994 /* Even if names are not encrypted, we still need to call
2995 * ll_fname_disk_to_usr in order to decode names as they are
2996 * coming from the wire.
2998 rc = llcrypt_fname_alloc_buffer(parent, NAME_MAX + 1, &lltr);
3006 rc = ll_fname_disk_to_usr(parent, 0, 0, &de_name,
3009 llcrypt_fname_free_buffer(&lltr);
3013 lltr.name[lltr.len] = '\0';
3015 if (lltr.len <= len_orig && gfpathlen >= lltr.len + 1) {
3016 memcpy(ptr, lltr.name, lltr.len);
3021 gfpathlen -= lltr.len + 1;
3025 llcrypt_fname_free_buffer(&lltr);
3027 if (rc == -EOVERFLOW) {
3034 /* We reached the end of the string, which means
3035 * we are dealing with the last component in the path.
3036 * So save a useless lookup and exit.
3042 if (enckey == 0 || nameenc == 0)
3046 de = lookup_one_len(p, de_parent, len);
3047 inode_unlock(parent);
3048 if (IS_ERR_OR_NULL(de) || !de->d_inode) {
3054 parent = de->d_inode;
3061 if (!IS_ERR_OR_NULL(de))
3066 int __ll_fid2path(struct inode *inode, struct getinfo_fid2path *gfout,
3067 size_t outsize, __u32 pathlen_orig)
3069 struct obd_export *exp = ll_i2mdexp(inode);
3072 /* Append root FID after gfout to let MDT know the root FID so that
3073 * it can lookup the correct path, this is mainly for fileset.
3074 * old server without fileset mount support will ignore this.
3076 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
3078 /* Call mdc_iocontrol */
3079 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
3081 if (!rc && gfout->gf_pathlen && gfout->gf_u.gf_path[0] == '/') {
3082 /* by convention, server side (mdt_path_current()) puts
3083 * a leading '/' to tell client that we are dealing with
3086 rc = fid2path_for_enc_file(inode, gfout->gf_u.gf_path,
3088 if (!rc && strlen(gfout->gf_u.gf_path) > pathlen_orig)
3095 int ll_fid2path(struct inode *inode, void __user *arg)
3097 const struct getinfo_fid2path __user *gfin = arg;
3098 __u32 pathlen, pathlen_orig;
3099 struct getinfo_fid2path *gfout;
3105 if (!capable(CAP_DAC_READ_SEARCH) &&
3106 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3109 /* Only need to get the buflen */
3110 if (get_user(pathlen, &gfin->gf_pathlen))
3113 if (pathlen > PATH_MAX)
3115 pathlen_orig = pathlen;
3118 outsize = sizeof(*gfout) + pathlen;
3119 OBD_ALLOC(gfout, outsize);
3123 if (copy_from_user(gfout, arg, sizeof(*gfout)))
3124 GOTO(gf_free, rc = -EFAULT);
3126 gfout->gf_pathlen = pathlen;
3127 rc = __ll_fid2path(inode, gfout, outsize, pathlen_orig);
3131 if (copy_to_user(arg, gfout, sizeof(*gfout) + pathlen_orig))
3135 OBD_FREE(gfout, outsize);
3136 if (rc == -ENAMETOOLONG) {
3137 pathlen += PATH_MAX;
3144 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
3146 struct cl_object *obj = ll_i2info(inode)->lli_clob;
3154 ioc->idv_version = 0;
3155 ioc->idv_layout_version = UINT_MAX;
3157 /* If no file object initialized, we consider its version is 0. */
3161 env = cl_env_get(&refcheck);
3163 RETURN(PTR_ERR(env));
3165 io = vvp_env_thread_io(env);
3167 io->u.ci_data_version.dv_data_version = 0;
3168 io->u.ci_data_version.dv_layout_version = UINT_MAX;
3169 io->u.ci_data_version.dv_flags = ioc->idv_flags;
3172 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
3173 result = cl_io_loop(env, io);
3175 result = io->ci_result;
3177 ioc->idv_version = io->u.ci_data_version.dv_data_version;
3178 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
3180 cl_io_fini(env, io);
3182 if (unlikely(io->ci_need_restart))
3185 cl_env_put(env, &refcheck);
3191 * Read the data_version for inode.
3193 * This value is computed using stripe object version on OST.
3194 * Version is computed using server side locking.
3196 * @param flags if do sync on the OST side;
3198 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
3199 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
3201 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
3203 struct ioc_data_version ioc = { .idv_flags = flags };
3206 rc = ll_ioc_data_version(inode, &ioc);
3208 *data_version = ioc.idv_version;
3214 * Trigger a HSM release request for the provided inode.
3216 int ll_hsm_release(struct inode *inode)
3219 struct obd_client_handle *och = NULL;
3220 __u64 data_version = 0;
3226 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
3227 ll_i2sbi(inode)->ll_fsname,
3228 PFID(&ll_i2info(inode)->lli_fid));
3230 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
3232 GOTO(out, rc = PTR_ERR(och));
3234 /* Grab latest data_version and [am]time values */
3235 rc = ll_data_version(inode, &data_version,
3236 LL_DV_WR_FLUSH | LL_DV_SZ_UPDATE);
3240 env = cl_env_get(&refcheck);
3242 GOTO(out, rc = PTR_ERR(env));
3244 rc = ll_merge_attr(env, inode);
3245 cl_env_put(env, &refcheck);
3247 /* If error happen, we have the wrong size for a file.
3253 /* Release the file.
3254 * NB: lease lock handle is released in mdc_hsm_release_pack() because
3255 * we still need it to pack l_remote_handle to MDT. */
3256 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
3262 if (och != NULL && !IS_ERR(och)) /* close the file */
3263 ll_lease_close(och, inode, NULL);
3268 struct ll_swap_stack {
3271 struct inode *inode1;
3272 struct inode *inode2;
3277 static int ll_swap_layouts(struct file *file1, struct file *file2,
3278 struct lustre_swap_layouts *lsl)
3280 struct mdc_swap_layouts msl;
3281 struct md_op_data *op_data;
3284 struct ll_swap_stack *llss = NULL;
3287 OBD_ALLOC_PTR(llss);
3291 llss->inode1 = file_inode(file1);
3292 llss->inode2 = file_inode(file2);
3294 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
3298 /* we use 2 bool because it is easier to swap than 2 bits */
3299 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
3300 llss->check_dv1 = true;
3302 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
3303 llss->check_dv2 = true;
3305 /* we cannot use lsl->sl_dvX directly because we may swap them */
3306 llss->dv1 = lsl->sl_dv1;
3307 llss->dv2 = lsl->sl_dv2;
3309 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
3310 if (rc == 0) /* same file, done! */
3313 if (rc < 0) { /* sequentialize it */
3314 swap(llss->inode1, llss->inode2);
3316 swap(llss->dv1, llss->dv2);
3317 swap(llss->check_dv1, llss->check_dv2);
3321 if (gid != 0) { /* application asks to flush dirty cache */
3322 rc = ll_get_grouplock(llss->inode1, file1, gid);
3326 rc = ll_get_grouplock(llss->inode2, file2, gid);
3328 ll_put_grouplock(llss->inode1, file1, gid);
3333 /* ultimate check, before swaping the layouts we check if
3334 * dataversion has changed (if requested) */
3335 if (llss->check_dv1) {
3336 rc = ll_data_version(llss->inode1, &dv, 0);
3339 if (dv != llss->dv1)
3340 GOTO(putgl, rc = -EAGAIN);
3343 if (llss->check_dv2) {
3344 rc = ll_data_version(llss->inode2, &dv, 0);
3347 if (dv != llss->dv2)
3348 GOTO(putgl, rc = -EAGAIN);
3351 /* struct md_op_data is used to send the swap args to the mdt
3352 * only flags is missing, so we use struct mdc_swap_layouts
3353 * through the md_op_data->op_data */
3354 /* flags from user space have to be converted before they are send to
3355 * server, no flag is sent today, they are only used on the client */
3358 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3359 0, LUSTRE_OPC_ANY, &msl);
3360 if (IS_ERR(op_data))
3361 GOTO(free, rc = PTR_ERR(op_data));
3363 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3364 sizeof(*op_data), op_data, NULL);
3365 ll_finish_md_op_data(op_data);
3372 ll_put_grouplock(llss->inode2, file2, gid);
3373 ll_put_grouplock(llss->inode1, file1, gid);
3383 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3385 struct obd_export *exp = ll_i2mdexp(inode);
3386 struct md_op_data *op_data;
3390 /* Detect out-of range masks */
3391 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3394 /* Non-root users are forbidden to set or clear flags which are
3395 * NOT defined in HSM_USER_MASK. */
3396 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3397 !capable(CAP_SYS_ADMIN))
3400 if (!exp_connect_archive_id_array(exp)) {
3401 /* Detect out-of range archive id */
3402 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3403 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3407 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3408 LUSTRE_OPC_ANY, hss);
3409 if (IS_ERR(op_data))
3410 RETURN(PTR_ERR(op_data));
3412 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3415 ll_finish_md_op_data(op_data);
3420 static int ll_hsm_import(struct inode *inode, struct file *file,
3421 struct hsm_user_import *hui)
3423 struct hsm_state_set *hss = NULL;
3424 struct iattr *attr = NULL;
3428 if (!S_ISREG(inode->i_mode))
3434 GOTO(out, rc = -ENOMEM);
3436 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3437 hss->hss_archive_id = hui->hui_archive_id;
3438 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3439 rc = ll_hsm_state_set(inode, hss);
3443 OBD_ALLOC_PTR(attr);
3445 GOTO(out, rc = -ENOMEM);
3447 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3448 attr->ia_mode |= S_IFREG;
3449 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3450 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3451 attr->ia_size = hui->hui_size;
3452 attr->ia_mtime.tv_sec = hui->hui_mtime;
3453 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3454 attr->ia_atime.tv_sec = hui->hui_atime;
3455 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3457 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3458 ATTR_UID | ATTR_GID |
3459 ATTR_MTIME | ATTR_MTIME_SET |
3460 ATTR_ATIME | ATTR_ATIME_SET;
3464 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3468 inode_unlock(inode);
3480 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3482 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3483 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3486 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3488 struct inode *inode = file_inode(file);
3490 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3491 ATTR_MTIME | ATTR_MTIME_SET |
3494 .tv_sec = lfu->lfu_atime_sec,
3495 .tv_nsec = lfu->lfu_atime_nsec,
3498 .tv_sec = lfu->lfu_mtime_sec,
3499 .tv_nsec = lfu->lfu_mtime_nsec,
3502 .tv_sec = lfu->lfu_ctime_sec,
3503 .tv_nsec = lfu->lfu_ctime_nsec,
3509 if (!capable(CAP_SYS_ADMIN))
3512 if (!S_ISREG(inode->i_mode))
3516 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3518 inode_unlock(inode);
3523 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3526 case MODE_READ_USER:
3528 case MODE_WRITE_USER:
3535 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3537 /* Used to allow the upper layers of the client to request an LDLM lock
3538 * without doing an actual read or write.
3540 * Used for ladvise lockahead to manually request specific locks.
3542 * \param[in] file file this ladvise lock request is on
3543 * \param[in] ladvise ladvise struct describing this lock request
3545 * \retval 0 success, no detailed result available (sync requests
3546 * and requests sent to the server [not handled locally]
3547 * cannot return detailed results)
3548 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3549 * see definitions for details.
3550 * \retval negative negative errno on error
3552 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3554 struct lu_env *env = NULL;
3555 struct cl_io *io = NULL;
3556 struct cl_lock *lock = NULL;
3557 struct cl_lock_descr *descr = NULL;
3558 struct dentry *dentry = file->f_path.dentry;
3559 struct inode *inode = dentry->d_inode;
3560 enum cl_lock_mode cl_mode;
3561 off_t start = ladvise->lla_start;
3562 off_t end = ladvise->lla_end;
3569 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3570 dentry, dentry->d_inode,
3571 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3574 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3576 GOTO(out, result = cl_mode);
3578 /* Get IO environment */
3579 result = cl_io_get(inode, &env, &io, &refcheck);
3583 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3586 * nothing to do for this io. This currently happens when
3587 * stripe sub-object's are not yet created.
3589 result = io->ci_result;
3590 } else if (result == 0) {
3591 lock = vvp_env_lock(env);
3592 descr = &lock->cll_descr;
3594 descr->cld_obj = io->ci_obj;
3595 /* Convert byte offsets to pages */
3596 descr->cld_start = cl_index(io->ci_obj, start);
3597 descr->cld_end = cl_index(io->ci_obj, end);
3598 descr->cld_mode = cl_mode;
3599 /* CEF_MUST is used because we do not want to convert a
3600 * lockahead request to a lockless lock */
3601 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3603 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3604 descr->cld_enq_flags |= CEF_SPECULATIVE;
3606 result = cl_lock_request(env, io, lock);
3608 /* On success, we need to release the lock */
3610 cl_lock_release(env, lock);
3612 cl_io_fini(env, io);
3613 cl_env_put(env, &refcheck);
3615 /* -ECANCELED indicates a matching lock with a different extent
3616 * was already present, and -EEXIST indicates a matching lock
3617 * on exactly the same extent was already present.
3618 * We convert them to positive values for userspace to make
3619 * recognizing true errors easier.
3620 * Note we can only return these detailed results on async requests,
3621 * as sync requests look the same as i/o requests for locking. */
3622 if (result == -ECANCELED)
3623 result = LLA_RESULT_DIFFERENT;
3624 else if (result == -EEXIST)
3625 result = LLA_RESULT_SAME;
3630 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3632 static int ll_ladvise_sanity(struct inode *inode,
3633 struct llapi_lu_ladvise *ladvise)
3635 struct ll_sb_info *sbi = ll_i2sbi(inode);
3636 enum lu_ladvise_type advice = ladvise->lla_advice;
3637 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3638 * be in the first 32 bits of enum ladvise_flags */
3639 __u32 flags = ladvise->lla_peradvice_flags;
3640 /* 3 lines at 80 characters per line, should be plenty */
3643 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3646 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3647 sbi->ll_fsname, advice,
3648 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3652 /* Per-advice checks */
3654 case LU_LADVISE_LOCKNOEXPAND:
3655 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3657 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3658 "rc = %d\n", sbi->ll_fsname, flags,
3659 ladvise_names[advice], rc);
3663 case LU_LADVISE_LOCKAHEAD:
3664 /* Currently only READ and WRITE modes can be requested */
3665 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3666 ladvise->lla_lockahead_mode == 0) {
3668 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3669 "rc = %d\n", sbi->ll_fsname,
3670 ladvise->lla_lockahead_mode,
3671 ladvise_names[advice], rc);
3675 case LU_LADVISE_WILLREAD:
3676 case LU_LADVISE_DONTNEED:
3678 /* Note fall through above - These checks apply to all advices
3679 * except LOCKNOEXPAND */
3680 if (flags & ~LF_DEFAULT_MASK) {
3682 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3683 "rc = %d\n", sbi->ll_fsname, flags,
3684 ladvise_names[advice], rc);
3687 if (ladvise->lla_start >= ladvise->lla_end) {
3689 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3690 "for %s: rc = %d\n", sbi->ll_fsname,
3691 ladvise->lla_start, ladvise->lla_end,
3692 ladvise_names[advice], rc);
3704 * Give file access advices
3706 * The ladvise interface is similar to Linux fadvise() system call, except it
3707 * forwards the advices directly from Lustre client to server. The server side
3708 * codes will apply appropriate read-ahead and caching techniques for the
3709 * corresponding files.
3711 * A typical workload for ladvise is e.g. a bunch of different clients are
3712 * doing small random reads of a file, so prefetching pages into OSS cache
3713 * with big linear reads before the random IO is a net benefit. Fetching
3714 * all that data into each client cache with fadvise() may not be, due to
3715 * much more data being sent to the client.
3717 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3718 struct llapi_lu_ladvise *ladvise)
3722 struct cl_ladvise_io *lio;
3727 env = cl_env_get(&refcheck);
3729 RETURN(PTR_ERR(env));
3731 io = vvp_env_thread_io(env);
3732 io->ci_obj = ll_i2info(inode)->lli_clob;
3734 /* initialize parameters for ladvise */
3735 lio = &io->u.ci_ladvise;
3736 lio->li_start = ladvise->lla_start;
3737 lio->li_end = ladvise->lla_end;
3738 lio->li_fid = ll_inode2fid(inode);
3739 lio->li_advice = ladvise->lla_advice;
3740 lio->li_flags = flags;
3742 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3743 rc = cl_io_loop(env, io);
3747 cl_io_fini(env, io);
3748 cl_env_put(env, &refcheck);
3752 static int ll_lock_noexpand(struct file *file, int flags)
3754 struct ll_file_data *fd = file->private_data;
3756 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3761 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3764 struct fsxattr fsxattr;
3766 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3769 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3770 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3771 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3772 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3773 if (copy_to_user(uarg, &fsxattr, sizeof(fsxattr)))
3779 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3783 * Project Quota ID state is only allowed to change from within the init
3784 * namespace. Enforce that restriction only if we are trying to change
3785 * the quota ID state. Everything else is allowed in user namespaces.
3787 if (current_user_ns() == &init_user_ns) {
3789 * Caller is allowed to change the project ID. if it is being
3790 * changed, make sure that the new value is valid.
3792 if (ll_i2info(inode)->lli_projid != projid &&
3793 !projid_valid(make_kprojid(&init_user_ns, projid)))
3799 if (ll_i2info(inode)->lli_projid != projid)
3802 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3803 if (!(xflags & FS_XFLAG_PROJINHERIT))
3806 if (xflags & FS_XFLAG_PROJINHERIT)
3813 static int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3815 struct ptlrpc_request *req = NULL;
3816 struct md_op_data *op_data;
3817 struct cl_object *obj;
3818 unsigned int inode_flags;
3821 CDEBUG(D_QUOTA, DFID" xflags=%x projid=%u\n",
3822 PFID(ll_inode2fid(inode)), xflags, projid);
3823 rc = ll_ioctl_check_project(inode, xflags, projid);
3827 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3828 LUSTRE_OPC_ANY, NULL);
3829 if (IS_ERR(op_data))
3830 RETURN(PTR_ERR(op_data));
3832 inode_flags = ll_xflags_to_inode_flags(xflags);
3833 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3834 if (xflags & FS_XFLAG_PROJINHERIT)
3835 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3837 /* pass projid to md_op_data */
3838 op_data->op_projid = projid;
3840 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3841 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3842 ptlrpc_req_finished(req);
3844 GOTO(out_fsxattr, rc);
3845 ll_update_inode_flags(inode, op_data->op_attr_flags);
3847 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3848 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3849 GOTO(out_fsxattr, rc);
3851 obj = ll_i2info(inode)->lli_clob;
3853 struct iattr attr = { 0 };
3855 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3859 ll_finish_md_op_data(op_data);
3863 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3866 struct fsxattr fsxattr;
3870 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3873 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3874 fsxattr.fsx_projid));
3877 int ll_ioctl_project(struct file *file, unsigned int cmd, void __user *uarg)
3879 struct lu_project lu_project;
3880 struct dentry *dentry = file_dentry(file);
3881 struct inode *inode = file_inode(file);
3882 struct dentry *child_dentry = NULL;
3883 int rc = 0, name_len;
3885 if (copy_from_user(&lu_project, uarg, sizeof(lu_project)))
3888 /* apply child dentry if name is valid */
3889 name_len = strnlen(lu_project.project_name, NAME_MAX);
3890 if (name_len > 0 && name_len <= NAME_MAX) {
3892 child_dentry = lookup_one_len(lu_project.project_name,
3894 inode_unlock(inode);
3895 if (IS_ERR(child_dentry)) {
3896 rc = PTR_ERR(child_dentry);
3899 inode = child_dentry->d_inode;
3904 } else if (name_len > NAME_MAX) {
3909 switch (lu_project.project_type) {
3910 case LU_PROJECT_SET:
3911 rc = ll_set_project(inode, lu_project.project_xflags,
3912 lu_project.project_id);
3914 case LU_PROJECT_GET:
3915 lu_project.project_xflags =
3916 ll_inode_flags_to_xflags(inode->i_flags);
3917 if (test_bit(LLIF_PROJECT_INHERIT,
3918 &ll_i2info(inode)->lli_flags))
3919 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3920 lu_project.project_id = ll_i2info(inode)->lli_projid;
3921 if (copy_to_user(uarg, &lu_project, sizeof(lu_project))) {
3931 if (!IS_ERR_OR_NULL(child_dentry))
3936 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3939 struct inode *inode = file_inode(file);
3940 struct ll_file_data *fd = file->private_data;
3941 struct ll_inode_info *lli = ll_i2info(inode);
3942 struct obd_client_handle *och = NULL;
3943 struct split_param sp;
3944 struct pcc_param param;
3945 bool lease_broken = false;
3947 enum mds_op_bias bias = 0;
3949 struct file *layout_file = NULL;
3951 size_t data_size = 0;
3952 bool attached = false;
3957 mutex_lock(&lli->lli_och_mutex);
3958 if (fd->fd_lease_och != NULL) {
3959 och = fd->fd_lease_och;
3960 fd->fd_lease_och = NULL;
3962 mutex_unlock(&lli->lli_och_mutex);
3967 fmode = och->och_flags;
3969 switch (ioc->lil_flags) {
3970 case LL_LEASE_RESYNC_DONE:
3971 if (ioc->lil_count > IOC_IDS_MAX)
3972 GOTO(out_lease_close, rc = -EINVAL);
3974 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3975 OBD_ALLOC(data, data_size);
3977 GOTO(out_lease_close, rc = -ENOMEM);
3979 if (copy_from_user(data, uarg, data_size))
3980 GOTO(out_lease_close, rc = -EFAULT);
3982 bias = MDS_CLOSE_RESYNC_DONE;
3984 case LL_LEASE_LAYOUT_MERGE: {
3985 if (ioc->lil_count != 1)
3986 GOTO(out_lease_close, rc = -EINVAL);
3988 uarg += sizeof(*ioc);
3989 if (copy_from_user(&fdv, uarg, sizeof(__u32)))
3990 GOTO(out_lease_close, rc = -EFAULT);
3992 layout_file = fget(fdv);
3994 GOTO(out_lease_close, rc = -EBADF);
3996 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3997 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3998 GOTO(out_lease_close, rc = -EPERM);
4000 data = file_inode(layout_file);
4001 bias = MDS_CLOSE_LAYOUT_MERGE;
4004 case LL_LEASE_LAYOUT_SPLIT: {
4007 if (ioc->lil_count != 2)
4008 GOTO(out_lease_close, rc = -EINVAL);
4010 uarg += sizeof(*ioc);
4011 if (copy_from_user(&fdv, uarg, sizeof(__u32)))
4012 GOTO(out_lease_close, rc = -EFAULT);
4014 uarg += sizeof(__u32);
4015 if (copy_from_user(&mirror_id, uarg, sizeof(__u32)))
4016 GOTO(out_lease_close, rc = -EFAULT);
4018 layout_file = fget(fdv);
4020 GOTO(out_lease_close, rc = -EBADF);
4022 /* if layout_file == file, it means to destroy the mirror */
4023 sp.sp_inode = file_inode(layout_file);
4024 sp.sp_mirror_id = (__u16)mirror_id;
4026 bias = MDS_CLOSE_LAYOUT_SPLIT;
4029 case LL_LEASE_PCC_ATTACH:
4030 if (ioc->lil_count != 1)
4033 if (IS_ENCRYPTED(inode))
4034 RETURN(-EOPNOTSUPP);
4036 uarg += sizeof(*ioc);
4037 if (copy_from_user(¶m.pa_archive_id, uarg, sizeof(__u32)))
4038 GOTO(out_lease_close, rc2 = -EFAULT);
4040 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
4042 GOTO(out_lease_close, rc2);
4045 /* Grab latest data version */
4046 rc2 = ll_data_version(inode, ¶m.pa_data_version,
4049 GOTO(out_lease_close, rc2);
4052 bias = MDS_PCC_ATTACH;
4055 /* without close intent */
4060 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
4064 rc = ll_lease_och_release(inode, file);
4073 if (ioc->lil_flags == LL_LEASE_RESYNC_DONE && data)
4074 OBD_FREE(data, data_size);
4079 if (ioc->lil_flags == LL_LEASE_PCC_ATTACH) {
4082 rc = pcc_readwrite_attach_fini(file, inode,
4083 param.pa_layout_gen,
4088 ll_layout_refresh(inode, &fd->fd_layout_version);
4091 rc = ll_lease_type_from_fmode(fmode);
4095 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
4098 struct inode *inode = file_inode(file);
4099 struct ll_inode_info *lli = ll_i2info(inode);
4100 struct ll_file_data *fd = file->private_data;
4101 struct obd_client_handle *och = NULL;
4102 __u64 open_flags = 0;
4108 switch (ioc->lil_mode) {
4109 case LL_LEASE_WRLCK:
4110 if (!(file->f_mode & FMODE_WRITE))
4112 fmode = FMODE_WRITE;
4114 case LL_LEASE_RDLCK:
4115 if (!(file->f_mode & FMODE_READ))
4119 case LL_LEASE_UNLCK:
4120 RETURN(ll_file_unlock_lease(file, ioc, uarg));
4125 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
4127 /* apply for lease */
4128 if (ioc->lil_flags & LL_LEASE_RESYNC)
4129 open_flags = MDS_OPEN_RESYNC;
4130 och = ll_lease_open(inode, file, fmode, open_flags);
4132 RETURN(PTR_ERR(och));
4134 if (ioc->lil_flags & LL_LEASE_RESYNC) {
4135 rc = ll_lease_file_resync(och, inode, uarg);
4137 ll_lease_close(och, inode, NULL);
4140 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
4142 ll_lease_close(och, inode, NULL);
4148 mutex_lock(&lli->lli_och_mutex);
4149 if (fd->fd_lease_och == NULL) {
4150 fd->fd_lease_och = och;
4153 mutex_unlock(&lli->lli_och_mutex);
4155 /* impossible now that only excl is supported for now */
4156 ll_lease_close(och, inode, &lease_broken);
4162 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
4164 struct ll_inode_info *lli = ll_i2info(inode);
4165 struct ll_sb_info *sbi = ll_i2sbi(inode);
4166 __u64 now = ktime_get_real_seconds();
4169 spin_lock(&lli->lli_heat_lock);
4170 heat->lh_flags = lli->lli_heat_flags;
4171 for (i = 0; i < heat->lh_count; i++)
4172 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
4173 now, sbi->ll_heat_decay_weight,
4174 sbi->ll_heat_period_second);
4175 spin_unlock(&lli->lli_heat_lock);
4178 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
4180 struct ll_inode_info *lli = ll_i2info(inode);
4183 spin_lock(&lli->lli_heat_lock);
4184 if (flags & LU_HEAT_FLAG_CLEAR)
4185 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
4187 if (flags & LU_HEAT_FLAG_OFF)
4188 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
4190 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
4192 spin_unlock(&lli->lli_heat_lock);
4198 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4200 struct inode *inode = file_inode(file);
4201 struct ll_file_data *fd = file->private_data;
4202 void __user *uarg = (void __user *)arg;
4206 CDEBUG(D_VFSTRACE|D_IOCTL, "VFS Op:inode="DFID"(%pK) cmd=%x arg=%lx\n",
4207 PFID(ll_inode2fid(inode)), inode, cmd, arg);
4208 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
4210 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
4211 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
4215 case LL_IOC_GETFLAGS:
4216 /* Get the current value of the file flags */
4217 return put_user(fd->fd_flags, (int __user *)arg);
4218 case LL_IOC_SETFLAGS:
4219 case LL_IOC_CLRFLAGS:
4220 /* Set or clear specific file flags */
4221 /* XXX This probably needs checks to ensure the flags are
4222 * not abused, and to handle any flag side effects.
4224 if (get_user(flags, (int __user *)arg))
4227 if (cmd == LL_IOC_SETFLAGS) {
4228 if ((flags & LL_FILE_IGNORE_LOCK) &&
4229 !(file->f_flags & O_DIRECT)) {
4231 CERROR("%s: unable to disable locking on non-O_DIRECT file "DFID": rc = %d\n",
4232 current->comm, PFID(ll_inode2fid(inode)),
4237 fd->fd_flags |= flags;
4239 fd->fd_flags &= ~flags;
4242 case LL_IOC_LOV_SETSTRIPE:
4243 case LL_IOC_LOV_SETSTRIPE_NEW:
4244 RETURN(ll_lov_setstripe(inode, file, uarg));
4245 case LL_IOC_LOV_SETEA:
4246 RETURN(ll_lov_setea(inode, file, uarg));
4247 case LL_IOC_LOV_SWAP_LAYOUTS: {
4249 struct lustre_swap_layouts lsl;
4251 if (copy_from_user(&lsl, uarg, sizeof(lsl)))
4254 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
4257 file2 = fget(lsl.sl_fd);
4261 /* O_WRONLY or O_RDWR */
4262 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
4263 GOTO(out, rc = -EPERM);
4265 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
4266 struct obd_client_handle *och = NULL;
4267 struct ll_inode_info *lli;
4268 struct inode *inode2;
4270 lli = ll_i2info(inode);
4271 mutex_lock(&lli->lli_och_mutex);
4272 if (fd->fd_lease_och != NULL) {
4273 och = fd->fd_lease_och;
4274 fd->fd_lease_och = NULL;
4276 mutex_unlock(&lli->lli_och_mutex);
4278 GOTO(out, rc = -ENOLCK);
4279 inode2 = file_inode(file2);
4280 rc = ll_swap_layouts_close(och, inode, inode2);
4282 rc = ll_swap_layouts(file, file2, &lsl);
4288 case LL_IOC_LOV_GETSTRIPE:
4289 case LL_IOC_LOV_GETSTRIPE_NEW:
4290 RETURN(ll_file_getstripe(inode, uarg, 0));
4291 case LL_IOC_GROUP_LOCK:
4292 RETURN(ll_get_grouplock(inode, file, arg));
4293 case LL_IOC_GROUP_UNLOCK:
4294 RETURN(ll_put_grouplock(inode, file, arg));
4295 case LL_IOC_DATA_VERSION: {
4296 struct ioc_data_version idv;
4299 if (copy_from_user(&idv, uarg, sizeof(idv)))
4302 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
4303 rc = ll_ioc_data_version(inode, &idv);
4305 if (rc == 0 && copy_to_user(uarg, &idv, sizeof(idv)))
4310 case LL_IOC_HSM_STATE_GET: {
4311 struct md_op_data *op_data;
4312 struct hsm_user_state *hus;
4319 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4320 LUSTRE_OPC_ANY, hus);
4321 if (IS_ERR(op_data)) {
4323 RETURN(PTR_ERR(op_data));
4326 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4329 if (copy_to_user(uarg, hus, sizeof(*hus)))
4332 ll_finish_md_op_data(op_data);
4336 case LL_IOC_HSM_STATE_SET: {
4337 struct hsm_state_set *hss;
4344 if (copy_from_user(hss, uarg, sizeof(*hss))) {
4349 rc = ll_hsm_state_set(inode, hss);
4354 case LL_IOC_HSM_ACTION: {
4355 struct md_op_data *op_data;
4356 struct hsm_current_action *hca;
4364 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4365 LUSTRE_OPC_ANY, hca);
4366 if (IS_ERR(op_data)) {
4368 RETURN(PTR_ERR(op_data));
4371 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4374 GOTO(skip_copy, rc);
4376 /* The hsm_current_action retreived from the server could
4377 * contain corrupt information. If it is incorrect data collect
4378 * debug information. We still send the data even if incorrect
4379 * to user land to handle.
4381 action = hsm_user_action2name(hca->hca_action);
4382 if (strcmp(action, "UNKNOWN") == 0 ||
4383 hca->hca_state > HPS_DONE) {
4385 "HSM current state %s action %s, offset = %llu, length %llu\n",
4386 hsm_progress_state2name(hca->hca_state), action,
4387 hca->hca_location.offset, hca->hca_location.length);
4390 if (copy_to_user(uarg, hca, sizeof(*hca)))
4393 ll_finish_md_op_data(op_data);
4397 case LL_IOC_SET_LEASE_OLD: {
4398 struct ll_ioc_lease ioc = { .lil_mode = arg };
4400 RETURN(ll_file_set_lease(file, &ioc, 0));
4402 case LL_IOC_SET_LEASE: {
4403 struct ll_ioc_lease ioc;
4405 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
4408 RETURN(ll_file_set_lease(file, &ioc, uarg));
4410 case LL_IOC_GET_LEASE: {
4411 struct ll_inode_info *lli = ll_i2info(inode);
4412 struct ldlm_lock *lock = NULL;
4415 mutex_lock(&lli->lli_och_mutex);
4416 if (fd->fd_lease_och != NULL) {
4417 struct obd_client_handle *och = fd->fd_lease_och;
4419 lock = ldlm_handle2lock(&och->och_lease_handle);
4421 lock_res_and_lock(lock);
4422 if (!ldlm_is_cancel(lock))
4423 fmode = och->och_flags;
4425 unlock_res_and_lock(lock);
4426 LDLM_LOCK_PUT(lock);
4429 mutex_unlock(&lli->lli_och_mutex);
4431 RETURN(ll_lease_type_from_fmode(fmode));
4433 case LL_IOC_HSM_IMPORT: {
4434 struct hsm_user_import *hui;
4440 if (copy_from_user(hui, uarg, sizeof(*hui))) {
4445 rc = ll_hsm_import(inode, file, hui);
4450 case LL_IOC_FUTIMES_3: {
4451 struct ll_futimes_3 lfu;
4453 if (copy_from_user(&lfu, uarg, sizeof(lfu)))
4456 RETURN(ll_file_futimes_3(file, &lfu));
4458 case LL_IOC_LADVISE: {
4459 struct llapi_ladvise_hdr *k_ladvise_hdr;
4460 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4463 int alloc_size = sizeof(*k_ladvise_hdr);
4466 u_ladvise_hdr = uarg;
4467 OBD_ALLOC_PTR(k_ladvise_hdr);
4468 if (k_ladvise_hdr == NULL)
4471 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4472 GOTO(out_ladvise, rc = -EFAULT);
4474 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4475 k_ladvise_hdr->lah_count < 1)
4476 GOTO(out_ladvise, rc = -EINVAL);
4478 num_advise = k_ladvise_hdr->lah_count;
4479 if (num_advise >= LAH_COUNT_MAX)
4480 GOTO(out_ladvise, rc = -EFBIG);
4482 OBD_FREE_PTR(k_ladvise_hdr);
4483 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4484 lah_advise[num_advise]);
4485 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4486 if (k_ladvise_hdr == NULL)
4490 * TODO: submit multiple advices to one server in a single RPC
4492 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4493 GOTO(out_ladvise, rc = -EFAULT);
4495 for (i = 0; i < num_advise; i++) {
4496 struct llapi_lu_ladvise *k_ladvise =
4497 &k_ladvise_hdr->lah_advise[i];
4498 struct llapi_lu_ladvise __user *u_ladvise =
4499 &u_ladvise_hdr->lah_advise[i];
4501 rc = ll_ladvise_sanity(inode, k_ladvise);
4503 GOTO(out_ladvise, rc);
4505 switch (k_ladvise->lla_advice) {
4506 case LU_LADVISE_LOCKNOEXPAND:
4507 rc = ll_lock_noexpand(file,
4508 k_ladvise->lla_peradvice_flags);
4509 GOTO(out_ladvise, rc);
4510 case LU_LADVISE_LOCKAHEAD:
4512 rc = ll_file_lock_ahead(file, k_ladvise);
4515 GOTO(out_ladvise, rc);
4518 &u_ladvise->lla_lockahead_result))
4519 GOTO(out_ladvise, rc = -EFAULT);
4522 rc = ll_ladvise(inode, file,
4523 k_ladvise_hdr->lah_flags,
4526 GOTO(out_ladvise, rc);
4533 OBD_FREE(k_ladvise_hdr, alloc_size);
4536 case LL_IOC_FLR_SET_MIRROR: {
4537 /* mirror I/O must be direct to avoid polluting page cache
4539 if (!(file->f_flags & O_DIRECT))
4542 fd->fd_designated_mirror = arg;
4545 case LL_IOC_HEAT_GET: {
4546 struct lu_heat uheat;
4547 struct lu_heat *heat;
4550 if (copy_from_user(&uheat, uarg, sizeof(uheat)))
4553 if (uheat.lh_count > OBD_HEAT_COUNT)
4554 uheat.lh_count = OBD_HEAT_COUNT;
4556 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4557 OBD_ALLOC(heat, size);
4561 heat->lh_count = uheat.lh_count;
4562 ll_heat_get(inode, heat);
4563 rc = copy_to_user(uarg, heat, size);
4564 OBD_FREE(heat, size);
4565 RETURN(rc ? -EFAULT : 0);
4567 case LL_IOC_HEAT_SET: {
4570 if (copy_from_user(&flags, uarg, sizeof(flags)))
4573 rc = ll_heat_set(inode, flags);
4576 case LL_IOC_PCC_DETACH: {
4577 struct lu_pcc_detach *detach;
4579 OBD_ALLOC_PTR(detach);
4583 if (copy_from_user(detach, uarg, sizeof(*detach)))
4584 GOTO(out_detach_free, rc = -EFAULT);
4586 if (!S_ISREG(inode->i_mode))
4587 GOTO(out_detach_free, rc = -EINVAL);
4589 if (!inode_owner_or_capable(&init_user_ns, inode))
4590 GOTO(out_detach_free, rc = -EPERM);
4592 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4594 OBD_FREE_PTR(detach);
4597 case LL_IOC_PCC_STATE: {
4598 struct lu_pcc_state __user *ustate = uarg;
4599 struct lu_pcc_state *state;
4601 OBD_ALLOC_PTR(state);
4605 if (copy_from_user(state, ustate, sizeof(*state)))
4606 GOTO(out_state, rc = -EFAULT);
4608 rc = pcc_ioctl_state(file, inode, state);
4610 GOTO(out_state, rc);
4612 if (copy_to_user(ustate, state, sizeof(*state)))
4613 GOTO(out_state, rc = -EFAULT);
4616 OBD_FREE_PTR(state);
4620 rc = ll_iocontrol(inode, file, cmd, uarg);
4623 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, uarg));
4627 loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4629 struct inode *inode = file_inode(file);
4632 struct cl_lseek_io *lsio;
4639 env = cl_env_get(&refcheck);
4641 RETURN(PTR_ERR(env));
4643 io = vvp_env_thread_io(env);
4644 io->ci_obj = ll_i2info(inode)->lli_clob;
4645 ll_io_set_mirror(io, file);
4647 lsio = &io->u.ci_lseek;
4648 lsio->ls_start = offset;
4649 lsio->ls_whence = whence;
4650 lsio->ls_result = -ENXIO;
4653 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4655 struct vvp_io *vio = vvp_env_io(env);
4657 vio->vui_fd = file->private_data;
4658 rc = cl_io_loop(env, io);
4662 retval = rc ? : lsio->ls_result;
4663 cl_io_fini(env, io);
4664 } while (unlikely(io->ci_need_restart));
4666 cl_env_put(env, &refcheck);
4668 /* Without the key, SEEK_HOLE return value has to be
4669 * rounded up to next LUSTRE_ENCRYPTION_UNIT_SIZE.
4671 if (llcrypt_require_key(inode) == -ENOKEY && whence == SEEK_HOLE)
4672 retval = round_up(retval, LUSTRE_ENCRYPTION_UNIT_SIZE);
4677 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4679 struct inode *inode = file_inode(file);
4680 loff_t retval = offset, eof = 0;
4681 ktime_t kstart = ktime_get();
4685 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4686 PFID(ll_inode2fid(inode)), inode, retval, retval,
4689 if (origin == SEEK_END) {
4690 retval = ll_glimpse_size(inode);
4693 eof = i_size_read(inode);
4696 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4700 /* flush local cache first if any */
4701 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4704 retval = ll_lseek(file, offset, origin);
4707 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4709 retval = generic_file_llseek_size(file, offset, origin,
4710 ll_file_maxbytes(inode), eof);
4713 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4714 ktime_us_delta(ktime_get(), kstart));
4718 static int ll_flush(struct file *file, fl_owner_t id)
4720 struct inode *inode = file_inode(file);
4721 struct ll_inode_info *lli = ll_i2info(inode);
4722 struct ll_file_data *fd = file->private_data;
4725 LASSERT(!S_ISDIR(inode->i_mode));
4727 /* catch async errors that were recorded back when async writeback
4728 * failed for pages in this mapping. */
4729 rc = lli->lli_async_rc;
4730 lli->lli_async_rc = 0;
4731 if (lli->lli_clob != NULL) {
4732 err = lov_read_and_clear_async_rc(lli->lli_clob);
4737 /* The application has been told write failure already.
4738 * Do not report failure again. */
4739 if (fd->fd_write_failed)
4741 return rc ? -EIO : 0;
4745 * Called to make sure a portion of file has been written out.
4746 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4748 * Return how many pages have been written.
4750 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4751 enum cl_fsync_mode mode, int ignore_layout)
4755 struct cl_fsync_io *fio;
4760 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4761 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4764 env = cl_env_get(&refcheck);
4766 RETURN(PTR_ERR(env));
4768 io = vvp_env_thread_io(env);
4769 io->ci_obj = ll_i2info(inode)->lli_clob;
4770 io->ci_ignore_layout = ignore_layout;
4772 /* initialize parameters for sync */
4773 fio = &io->u.ci_fsync;
4774 fio->fi_start = start;
4776 fio->fi_fid = ll_inode2fid(inode);
4777 fio->fi_mode = mode;
4778 fio->fi_nr_written = 0;
4780 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4781 result = cl_io_loop(env, io);
4783 result = io->ci_result;
4785 result = fio->fi_nr_written;
4786 cl_io_fini(env, io);
4787 cl_env_put(env, &refcheck);
4793 * When dentry is provided (the 'else' case), file_dentry() may be
4794 * null and dentry must be used directly rather than pulled from
4795 * file_dentry() as is done otherwise.
4798 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4800 struct dentry *dentry = file_dentry(file);
4801 struct inode *inode = dentry->d_inode;
4802 struct ll_inode_info *lli = ll_i2info(inode);
4803 struct ptlrpc_request *req;
4804 ktime_t kstart = ktime_get();
4810 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4811 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4813 /* fsync's caller has already called _fdata{sync,write}, we want
4814 * that IO to finish before calling the osc and mdc sync methods */
4815 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4817 /* catch async errors that were recorded back when async writeback
4818 * failed for pages in this mapping. */
4819 if (!S_ISDIR(inode->i_mode)) {
4820 err = lli->lli_async_rc;
4821 lli->lli_async_rc = 0;
4824 if (lli->lli_clob != NULL) {
4825 err = lov_read_and_clear_async_rc(lli->lli_clob);
4831 if (S_ISREG(inode->i_mode) && !lli->lli_synced_to_mds) {
4833 * only the first sync on MDS makes sense,
4834 * everything else is stored on OSTs
4836 err = md_fsync(ll_i2sbi(inode)->ll_md_exp,
4837 ll_inode2fid(inode), &req);
4841 lli->lli_synced_to_mds = true;
4842 ptlrpc_req_finished(req);
4846 if (S_ISREG(inode->i_mode)) {
4847 struct ll_file_data *fd = file->private_data;
4850 /* Sync metadata on MDT first, and then sync the cached data
4853 err = pcc_fsync(file, start, end, datasync, &cached);
4855 err = cl_sync_file_range(inode, start, end,
4857 if (rc == 0 && err < 0)
4860 fd->fd_write_failed = true;
4862 fd->fd_write_failed = false;
4866 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4867 ktime_us_delta(ktime_get(), kstart));
4872 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4874 struct inode *inode = file_inode(file);
4875 struct ll_sb_info *sbi = ll_i2sbi(inode);
4876 struct ldlm_enqueue_info einfo = {
4877 .ei_type = LDLM_FLOCK,
4878 .ei_cb_cp = ldlm_flock_completion_ast,
4879 .ei_cbdata = file_lock,
4881 struct md_op_data *op_data;
4882 struct lustre_handle lockh = { 0 };
4883 union ldlm_policy_data flock = { { 0 } };
4884 int fl_type = file_lock->fl_type;
4885 ktime_t kstart = ktime_get();
4891 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4892 PFID(ll_inode2fid(inode)), file_lock);
4894 if (file_lock->fl_flags & FL_FLOCK) {
4895 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4896 /* flocks are whole-file locks */
4897 flock.l_flock.end = OFFSET_MAX;
4898 /* For flocks owner is determined by the local file desctiptor*/
4899 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4900 } else if (file_lock->fl_flags & FL_POSIX) {
4901 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4902 flock.l_flock.start = file_lock->fl_start;
4903 flock.l_flock.end = file_lock->fl_end;
4907 flock.l_flock.pid = file_lock->fl_pid;
4909 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4910 /* Somewhat ugly workaround for svc lockd.
4911 * lockd installs custom fl_lmops->lm_compare_owner that checks
4912 * for the fl_owner to be the same (which it always is on local node
4913 * I guess between lockd processes) and then compares pid.
4914 * As such we assign pid to the owner field to make it all work,
4915 * conflict with normal locks is unlikely since pid space and
4916 * pointer space for current->files are not intersecting */
4917 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4918 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4923 einfo.ei_mode = LCK_PR;
4926 /* An unlock request may or may not have any relation to
4927 * existing locks so we may not be able to pass a lock handle
4928 * via a normal ldlm_lock_cancel() request. The request may even
4929 * unlock a byte range in the middle of an existing lock. In
4930 * order to process an unlock request we need all of the same
4931 * information that is given with a normal read or write record
4932 * lock request. To avoid creating another ldlm unlock (cancel)
4933 * message we'll treat a LCK_NL flock request as an unlock. */
4934 einfo.ei_mode = LCK_NL;
4937 einfo.ei_mode = LCK_PW;
4941 CERROR("%s: fcntl from '%s' unknown lock type=%d: rc = %d\n",
4942 sbi->ll_fsname, current->comm, fl_type, rc);
4957 flags = LDLM_FL_BLOCK_NOWAIT;
4963 flags = LDLM_FL_TEST_LOCK;
4967 CERROR("%s: fcntl from '%s' unknown lock command=%d: rc = %d\n",
4968 sbi->ll_fsname, current->comm, cmd, rc);
4972 /* Save the old mode so that if the mode in the lock changes we
4973 * can decrement the appropriate reader or writer refcount. */
4974 file_lock->fl_type = einfo.ei_mode;
4976 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4977 LUSTRE_OPC_ANY, NULL);
4978 if (IS_ERR(op_data))
4979 RETURN(PTR_ERR(op_data));
4981 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4982 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4983 flock.l_flock.pid, flags, einfo.ei_mode,
4984 flock.l_flock.start, flock.l_flock.end);
4986 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4989 /* Restore the file lock type if not TEST lock. */
4990 if (!(flags & LDLM_FL_TEST_LOCK))
4991 file_lock->fl_type = fl_type;
4993 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4994 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4995 !(flags & LDLM_FL_TEST_LOCK))
4996 rc2 = locks_lock_file_wait(file, file_lock);
4998 if ((file_lock->fl_flags & FL_FLOCK) &&
4999 (rc == 0 || file_lock->fl_type == F_UNLCK))
5000 rc2 = flock_lock_file_wait(file, file_lock);
5001 if ((file_lock->fl_flags & FL_POSIX) &&
5002 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
5003 !(flags & LDLM_FL_TEST_LOCK))
5004 rc2 = posix_lock_file_wait(file, file_lock);
5005 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
5007 if (rc2 && file_lock->fl_type != F_UNLCK) {
5008 einfo.ei_mode = LCK_NL;
5009 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
5014 ll_finish_md_op_data(op_data);
5017 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
5018 ktime_us_delta(ktime_get(), kstart));
5022 int ll_get_fid_by_name(struct inode *parent, const char *name,
5023 int namelen, struct lu_fid *fid,
5024 struct inode **inode)
5026 struct md_op_data *op_data = NULL;
5027 struct mdt_body *body;
5028 struct ptlrpc_request *req;
5032 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
5033 LUSTRE_OPC_ANY, NULL);
5034 if (IS_ERR(op_data))
5035 RETURN(PTR_ERR(op_data));
5037 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
5038 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
5039 ll_finish_md_op_data(op_data);
5043 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
5045 GOTO(out_req, rc = -EFAULT);
5047 *fid = body->mbo_fid1;
5050 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
5052 ptlrpc_req_finished(req);
5056 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
5057 const char *name, __u32 flags)
5059 struct dentry *dchild = NULL;
5060 struct inode *child_inode = NULL;
5061 struct md_op_data *op_data;
5062 struct ptlrpc_request *request = NULL;
5063 struct obd_client_handle *och = NULL;
5065 struct mdt_body *body;
5066 __u64 data_version = 0;
5067 size_t namelen = strlen(name);
5068 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
5072 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
5073 PFID(ll_inode2fid(parent)), name,
5074 lum->lum_stripe_offset, lum->lum_stripe_count);
5076 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
5077 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
5078 lustre_swab_lmv_user_md(lum);
5080 /* Get child FID first */
5081 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
5084 dchild = d_lookup(file_dentry(file), &qstr);
5086 if (dchild->d_inode)
5087 child_inode = igrab(dchild->d_inode);
5092 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
5101 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
5102 OBD_CONNECT2_DIR_MIGRATE)) {
5103 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
5104 ll_dir_striped(child_inode)) {
5105 CERROR("%s: MDT doesn't support stripe directory "
5106 "migration!\n", ll_i2sbi(parent)->ll_fsname);
5107 GOTO(out_iput, rc = -EOPNOTSUPP);
5112 * lfs migrate command needs to be blocked on the client
5113 * by checking the migrate FID against the FID of the
5116 if (is_root_inode(child_inode))
5117 GOTO(out_iput, rc = -EINVAL);
5119 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
5120 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
5121 if (IS_ERR(op_data))
5122 GOTO(out_iput, rc = PTR_ERR(op_data));
5124 inode_lock(child_inode);
5125 op_data->op_fid3 = *ll_inode2fid(child_inode);
5126 if (!fid_is_sane(&op_data->op_fid3)) {
5127 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
5128 ll_i2sbi(parent)->ll_fsname, name,
5129 PFID(&op_data->op_fid3));
5130 GOTO(out_unlock, rc = -EINVAL);
5133 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
5134 op_data->op_data = lum;
5135 op_data->op_data_size = lumlen;
5137 /* migrate dirent only for subdirs if MDS_MIGRATE_NSONLY set */
5138 if (S_ISDIR(child_inode->i_mode) && (flags & MDS_MIGRATE_NSONLY) &&
5139 lmv_dir_layout_changing(ll_i2info(parent)->lli_lsm_md))
5140 op_data->op_bias |= MDS_MIGRATE_NSONLY;
5143 if (S_ISREG(child_inode->i_mode)) {
5144 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
5148 GOTO(out_unlock, rc);
5151 rc = ll_data_version(child_inode, &data_version,
5154 GOTO(out_close, rc);
5156 op_data->op_open_handle = och->och_open_handle;
5157 op_data->op_data_version = data_version;
5158 op_data->op_lease_handle = och->och_lease_handle;
5159 op_data->op_bias |= MDS_CLOSE_MIGRATE;
5161 spin_lock(&och->och_mod->mod_open_req->rq_lock);
5162 och->och_mod->mod_open_req->rq_replay = 0;
5163 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
5166 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
5167 op_data->op_name, op_data->op_namelen,
5168 op_data->op_name, op_data->op_namelen, &request);
5170 LASSERT(request != NULL);
5171 ll_update_times(request, parent);
5174 if (rc == 0 || rc == -EAGAIN) {
5175 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
5176 LASSERT(body != NULL);
5178 /* If the server does release layout lock, then we cleanup
5179 * the client och here, otherwise release it in out_close: */
5180 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
5181 obd_mod_put(och->och_mod);
5182 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
5184 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
5190 if (request != NULL) {
5191 ptlrpc_req_finished(request);
5195 /* Try again if the lease has cancelled. */
5196 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
5201 ll_lease_close(och, child_inode, NULL);
5203 clear_nlink(child_inode);
5205 inode_unlock(child_inode);
5206 ll_finish_md_op_data(op_data);
5213 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
5215 struct ll_file_data *fd = file->private_data;
5219 * In order to avoid flood of warning messages, only print one message
5220 * for one file. And the entire message rate on the client is limited
5221 * by CDEBUG_LIMIT too.
5223 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
5224 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
5225 CDEBUG_LIMIT(D_CONSOLE,
5226 "flock disabled, mount with '-o [local]flock' to enable\r\n");
5232 * test if some locks matching bits and l_req_mode are acquired
5233 * - bits can be in different locks
5234 * - if found clear the common lock bits in *bits
5235 * - the bits not found, are kept in *bits
5237 * \param bits [IN] searched lock bits [IN]
5238 * \param l_req_mode [IN] searched lock mode
5239 * \retval boolean, true iff all bits are found
5241 int ll_have_md_lock(struct obd_export *exp, struct inode *inode, __u64 *bits,
5242 enum ldlm_mode l_req_mode)
5244 struct lustre_handle lockh;
5245 union ldlm_policy_data policy;
5246 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
5247 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5256 fid = &ll_i2info(inode)->lli_fid;
5257 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5258 ldlm_lockname[mode]);
5260 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5261 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5262 policy.l_inodebits.bits = *bits & BIT(i);
5263 if (policy.l_inodebits.bits == 0)
5266 if (md_lock_match(exp, flags, fid, LDLM_IBITS, &policy, mode,
5268 struct ldlm_lock *lock;
5270 lock = ldlm_handle2lock(&lockh);
5273 ~(lock->l_policy_data.l_inodebits.bits);
5274 LDLM_LOCK_PUT(lock);
5276 *bits &= ~policy.l_inodebits.bits;
5283 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5284 struct lustre_handle *lockh, __u64 flags,
5285 enum ldlm_mode mode)
5287 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5292 fid = &ll_i2info(inode)->lli_fid;
5293 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5295 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5296 fid, LDLM_IBITS, &policy, mode, lockh);
5301 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5303 /* Already unlinked. Just update nlink and return success */
5304 if (rc == -ENOENT) {
5306 /* If it is striped directory, and there is bad stripe
5307 * Let's revalidate the dentry again, instead of returning
5309 if (ll_dir_striped(inode))
5312 /* This path cannot be hit for regular files unless in
5313 * case of obscure races, so no need to to validate
5315 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5317 } else if (rc != 0) {
5318 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5319 "%s: revalidate FID "DFID" error: rc = %d\n",
5320 ll_i2sbi(inode)->ll_fsname,
5321 PFID(ll_inode2fid(inode)), rc);
5327 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5329 struct inode *parent;
5330 struct inode *inode = dentry->d_inode;
5331 struct obd_export *exp = ll_i2mdexp(inode);
5332 struct lookup_intent oit = {
5335 struct ptlrpc_request *req = NULL;
5336 struct md_op_data *op_data;
5337 const char *name = NULL;
5342 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5343 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5345 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5346 parent = dentry->d_parent->d_inode;
5347 name = dentry->d_name.name;
5348 namelen = dentry->d_name.len;
5353 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5354 LUSTRE_OPC_ANY, NULL);
5355 if (IS_ERR(op_data))
5356 RETURN(PTR_ERR(op_data));
5358 /* Call getattr by fid */
5359 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5360 op_data->op_flags = MF_GETATTR_BY_FID;
5361 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5362 ll_finish_md_op_data(op_data);
5364 rc = ll_inode_revalidate_fini(inode, rc);
5368 rc = ll_revalidate_it_finish(req, &oit, dentry);
5370 ll_intent_release(&oit);
5374 /* Unlinked? Unhash dentry, so it is not picked up later by
5375 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5376 * here to preserve get_cwd functionality on 2.6.
5378 if (!dentry->d_inode->i_nlink)
5379 d_lustre_invalidate(dentry);
5381 ll_lookup_finish_locks(&oit, dentry);
5383 ptlrpc_req_finished(req);
5388 static int ll_merge_md_attr(struct inode *inode)
5390 struct ll_inode_info *lli = ll_i2info(inode);
5391 struct cl_attr attr = { 0 };
5394 if (!lli->lli_lsm_md)
5397 down_read(&lli->lli_lsm_sem);
5398 if (!lmv_dir_striped(lli->lli_lsm_md)) {
5399 up_read(&lli->lli_lsm_sem);
5402 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
5403 &attr, ll_md_blocking_ast);
5404 up_read(&lli->lli_lsm_sem);
5408 spin_lock(&inode->i_lock);
5409 set_nlink(inode, attr.cat_nlink);
5410 spin_unlock(&inode->i_lock);
5412 inode->i_blocks = attr.cat_blocks;
5413 i_size_write(inode, attr.cat_size);
5415 ll_i2info(inode)->lli_atime = attr.cat_atime;
5416 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5417 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5422 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5423 unsigned int flags, bool foreign)
5425 struct inode *inode = de->d_inode;
5426 struct ll_sb_info *sbi = ll_i2sbi(inode);
5427 struct ll_inode_info *lli = ll_i2info(inode);
5428 struct inode *dir = de->d_parent->d_inode;
5429 bool need_glimpse = true;
5430 ktime_t kstart = ktime_get();
5433 /* The OST object(s) determine the file size, blocks and mtime. */
5434 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5435 request_mask & STATX_MTIME))
5436 need_glimpse = false;
5438 if (dentry_may_statahead(dir, de))
5439 ll_start_statahead(dir, de, need_glimpse &&
5440 !(flags & AT_STATX_DONT_SYNC));
5442 if (flags & AT_STATX_DONT_SYNC)
5443 GOTO(fill_attr, rc = 0);
5445 rc = ll_inode_revalidate(de, IT_GETATTR);
5449 /* foreign file/dir are always of zero length, so don't
5450 * need to validate size.
5452 if (S_ISREG(inode->i_mode) && !foreign) {
5456 GOTO(fill_attr, rc);
5458 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5459 if (cached && rc < 0)
5463 GOTO(fill_attr, rc);
5466 * If the returned attr is masked with OBD_MD_FLSIZE &
5467 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5468 * or blocks obtained from MDT is strictly correct, and the file
5469 * is usually not being modified by clients, and the [a|m|c]time
5470 * got from MDT is also strictly correct.
5471 * Under this circumstance, it does not need to send glimpse
5472 * RPCs to OSTs for file attributes such as the size and blocks.
5474 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5475 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5476 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5477 inode->i_mtime.tv_sec = lli->lli_mtime;
5478 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5479 inode->i_atime.tv_sec = lli->lli_atime;
5480 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5481 inode->i_ctime.tv_sec = lli->lli_ctime;
5482 GOTO(fill_attr, rc);
5485 /* In case of restore, the MDT has the right size and has
5486 * already send it back without granting the layout lock,
5487 * inode is up-to-date so glimpse is useless.
5488 * Also to glimpse we need the layout, in case of a running
5489 * restore the MDT holds the layout lock so the glimpse will
5490 * block up to the end of restore (getattr will block)
5492 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5493 rc = ll_glimpse_size(inode);
5498 /* If object isn't regular a file then don't validate size. */
5499 /* foreign dir is not striped dir */
5500 if (ll_dir_striped(inode) && !foreign) {
5501 rc = ll_merge_md_attr(inode);
5506 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5507 inode->i_atime.tv_sec = lli->lli_atime;
5508 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5509 inode->i_mtime.tv_sec = lli->lli_mtime;
5510 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5511 inode->i_ctime.tv_sec = lli->lli_ctime;
5515 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5517 if (ll_need_32bit_api(sbi)) {
5518 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5519 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5520 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5522 stat->ino = inode->i_ino;
5523 stat->dev = inode->i_sb->s_dev;
5524 stat->rdev = inode->i_rdev;
5527 /* foreign symlink to be exposed as a real symlink */
5529 stat->mode = inode->i_mode;
5531 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5533 stat->uid = inode->i_uid;
5534 stat->gid = inode->i_gid;
5535 stat->atime = inode->i_atime;
5536 stat->mtime = inode->i_mtime;
5537 stat->ctime = inode->i_ctime;
5538 /* stat->blksize is used to tell about preferred IO size */
5539 if (sbi->ll_stat_blksize)
5540 stat->blksize = sbi->ll_stat_blksize;
5541 else if (S_ISREG(inode->i_mode))
5542 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
5543 LL_MAX_BLKSIZE_BITS);
5545 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5547 stat->nlink = inode->i_nlink;
5548 stat->size = i_size_read(inode);
5549 stat->blocks = inode->i_blocks;
5551 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5552 if (flags & AT_STATX_DONT_SYNC) {
5553 if (stat->size == 0 &&
5554 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5555 stat->size = lli->lli_lazysize;
5556 if (stat->blocks == 0 &&
5557 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5558 stat->blocks = lli->lli_lazyblocks;
5561 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5562 stat->result_mask |= STATX_BTIME;
5563 stat->btime.tv_sec = lli->lli_btime;
5566 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5567 #ifdef HAVE_LUSTRE_CRYPTO
5568 stat->attributes_mask |= STATX_ATTR_ENCRYPTED;
5570 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5571 /* if Lustre specific LUSTRE_ENCRYPT_FL flag is set, also set
5572 * ext4 equivalent to please statx
5574 if (stat->attributes & LUSTRE_ENCRYPT_FL)
5575 stat->attributes |= STATX_ATTR_ENCRYPTED;
5576 stat->result_mask &= request_mask;
5579 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5580 ktime_us_delta(ktime_get(), kstart));
5585 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5586 int ll_getattr(struct user_namespace *mnt_userns, const struct path *path,
5587 struct kstat *stat, u32 request_mask, unsigned int flags)
5589 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5593 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5595 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5596 AT_STATX_SYNC_AS_STAT, false);
5600 int cl_falloc(struct file *file, struct inode *inode, int mode, loff_t offset,
5603 loff_t size = i_size_read(inode);
5611 env = cl_env_get(&refcheck);
5613 RETURN(PTR_ERR(env));
5615 io = vvp_env_thread_io(env);
5616 io->ci_obj = ll_i2info(inode)->lli_clob;
5617 ll_io_set_mirror(io, file);
5619 io->ci_verify_layout = 1;
5620 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5621 io->u.ci_setattr.sa_falloc_mode = mode;
5622 io->u.ci_setattr.sa_falloc_offset = offset;
5623 io->u.ci_setattr.sa_falloc_end = offset + len;
5624 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5626 CDEBUG(D_INODE, "UID %u GID %u PRJID %u\n",
5627 from_kuid(&init_user_ns, inode->i_uid),
5628 from_kgid(&init_user_ns, inode->i_gid),
5629 ll_i2info(inode)->lli_projid);
5631 io->u.ci_setattr.sa_falloc_uid = from_kuid(&init_user_ns, inode->i_uid);
5632 io->u.ci_setattr.sa_falloc_gid = from_kgid(&init_user_ns, inode->i_gid);
5633 io->u.ci_setattr.sa_falloc_projid = ll_i2info(inode)->lli_projid;
5635 if (io->u.ci_setattr.sa_falloc_end > size) {
5636 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5638 /* Check new size against VFS/VM file size limit and rlimit */
5639 rc = inode_newsize_ok(inode, newsize);
5642 if (newsize > ll_file_maxbytes(inode)) {
5643 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5644 (unsigned long long)newsize,
5645 ll_file_maxbytes(inode));
5652 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5654 rc = cl_io_loop(env, io);
5657 cl_io_fini(env, io);
5658 } while (unlikely(io->ci_need_restart));
5661 cl_env_put(env, &refcheck);
5665 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5667 struct inode *inode = file_inode(filp);
5670 if (offset < 0 || len <= 0)
5673 * Encrypted inodes can't handle collapse range or zero range or insert
5674 * range since we would need to re-encrypt blocks with a different IV or
5675 * XTS tweak (which are based on the logical block number).
5676 * Similar to what ext4 does.
5678 if (IS_ENCRYPTED(inode) &&
5679 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5680 FALLOC_FL_ZERO_RANGE)))
5681 RETURN(-EOPNOTSUPP);
5684 * mode == 0 (which is standard prealloc) and PUNCH is supported
5685 * Rest of mode options are not supported yet.
5687 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5688 RETURN(-EOPNOTSUPP);
5690 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5692 rc = cl_falloc(filp, inode, mode, offset, len);
5694 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5695 * used by Lustre in several places. Retuning it here would
5696 * confuse applications that explicity test for EOPNOTSUPP
5697 * (95) and fall back to ftruncate().
5699 if (rc == -ENOTSUPP)
5705 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5706 __u64 start, __u64 len)
5710 struct fiemap *fiemap;
5711 unsigned int extent_count = fieinfo->fi_extents_max;
5713 num_bytes = sizeof(*fiemap) + (extent_count *
5714 sizeof(struct fiemap_extent));
5715 OBD_ALLOC_LARGE(fiemap, num_bytes);
5720 fiemap->fm_flags = fieinfo->fi_flags;
5721 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5722 fiemap->fm_start = start;
5723 fiemap->fm_length = len;
5724 if (extent_count > 0 &&
5725 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5726 sizeof(struct fiemap_extent)) != 0)
5727 GOTO(out, rc = -EFAULT);
5729 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5731 if (IS_ENCRYPTED(inode)) {
5734 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5735 fiemap->fm_extents[i].fe_flags |=
5736 FIEMAP_EXTENT_DATA_ENCRYPTED |
5737 FIEMAP_EXTENT_ENCODED;
5740 fieinfo->fi_flags = fiemap->fm_flags;
5741 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5742 if (extent_count > 0 &&
5743 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5744 fiemap->fm_mapped_extents *
5745 sizeof(struct fiemap_extent)) != 0)
5746 GOTO(out, rc = -EFAULT);
5748 OBD_FREE_LARGE(fiemap, num_bytes);
5752 int ll_inode_permission(struct user_namespace *mnt_userns, struct inode *inode,
5756 struct ll_sb_info *sbi;
5757 struct root_squash_info *squash;
5758 struct cred *cred = NULL;
5759 const struct cred *old_cred = NULL;
5760 bool squash_id = false;
5761 ktime_t kstart = ktime_get();
5765 if (mask & MAY_NOT_BLOCK)
5769 * as root inode are NOT getting validated in lookup operation,
5770 * need to revalidate PERM before permission check.
5772 if (is_root_inode(inode)) {
5773 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_GETATTR);
5778 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5779 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5781 /* squash fsuid/fsgid if needed */
5782 sbi = ll_i2sbi(inode);
5783 squash = &sbi->ll_squash;
5784 if (unlikely(squash->rsi_uid != 0 &&
5785 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5786 !test_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags))) {
5790 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5791 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5792 squash->rsi_uid, squash->rsi_gid);
5794 /* update current process's credentials
5795 * and FS capability */
5796 cred = prepare_creds();
5800 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5801 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5802 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5803 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5805 old_cred = override_creds(cred);
5808 rc = generic_permission(mnt_userns, inode, mask);
5809 /* restore current process's credentials and FS capability */
5811 revert_creds(old_cred);
5816 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5817 ktime_us_delta(ktime_get(), kstart));
5822 /* -o localflock - only provides locally consistent flock locks */
5823 static const struct file_operations ll_file_operations = {
5824 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5825 # ifdef HAVE_SYNC_READ_WRITE
5826 .read = new_sync_read,
5827 .write = new_sync_write,
5829 .read_iter = ll_file_read_iter,
5830 .write_iter = ll_file_write_iter,
5831 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5832 .read = ll_file_read,
5833 .aio_read = ll_file_aio_read,
5834 .write = ll_file_write,
5835 .aio_write = ll_file_aio_write,
5836 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5837 .unlocked_ioctl = ll_file_ioctl,
5838 .open = ll_file_open,
5839 .release = ll_file_release,
5840 .mmap = ll_file_mmap,
5841 .llseek = ll_file_seek,
5842 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5843 .splice_read = generic_file_splice_read,
5845 .splice_read = pcc_file_splice_read,
5847 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5848 .splice_write = iter_file_splice_write,
5852 .fallocate = ll_fallocate,
5855 static const struct file_operations ll_file_operations_flock = {
5856 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5857 # ifdef HAVE_SYNC_READ_WRITE
5858 .read = new_sync_read,
5859 .write = new_sync_write,
5860 # endif /* HAVE_SYNC_READ_WRITE */
5861 .read_iter = ll_file_read_iter,
5862 .write_iter = ll_file_write_iter,
5863 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5864 .read = ll_file_read,
5865 .aio_read = ll_file_aio_read,
5866 .write = ll_file_write,
5867 .aio_write = ll_file_aio_write,
5868 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5869 .unlocked_ioctl = ll_file_ioctl,
5870 .open = ll_file_open,
5871 .release = ll_file_release,
5872 .mmap = ll_file_mmap,
5873 .llseek = ll_file_seek,
5874 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5875 .splice_read = generic_file_splice_read,
5877 .splice_read = pcc_file_splice_read,
5879 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5880 .splice_write = iter_file_splice_write,
5884 .flock = ll_file_flock,
5885 .lock = ll_file_flock,
5886 .fallocate = ll_fallocate,
5889 /* These are for -o noflock - to return ENOSYS on flock calls */
5890 static const struct file_operations ll_file_operations_noflock = {
5891 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5892 # ifdef HAVE_SYNC_READ_WRITE
5893 .read = new_sync_read,
5894 .write = new_sync_write,
5895 # endif /* HAVE_SYNC_READ_WRITE */
5896 .read_iter = ll_file_read_iter,
5897 .write_iter = ll_file_write_iter,
5898 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5899 .read = ll_file_read,
5900 .aio_read = ll_file_aio_read,
5901 .write = ll_file_write,
5902 .aio_write = ll_file_aio_write,
5903 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5904 .unlocked_ioctl = ll_file_ioctl,
5905 .open = ll_file_open,
5906 .release = ll_file_release,
5907 .mmap = ll_file_mmap,
5908 .llseek = ll_file_seek,
5909 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5910 .splice_read = generic_file_splice_read,
5912 .splice_read = pcc_file_splice_read,
5914 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5915 .splice_write = iter_file_splice_write,
5919 .flock = ll_file_noflock,
5920 .lock = ll_file_noflock,
5921 .fallocate = ll_fallocate,
5924 const struct inode_operations ll_file_inode_operations = {
5925 .setattr = ll_setattr,
5926 .getattr = ll_getattr,
5927 .permission = ll_inode_permission,
5928 #ifdef HAVE_IOP_XATTR
5929 .setxattr = ll_setxattr,
5930 .getxattr = ll_getxattr,
5931 .removexattr = ll_removexattr,
5933 .listxattr = ll_listxattr,
5934 .fiemap = ll_fiemap,
5935 .get_acl = ll_get_acl,
5936 #ifdef HAVE_IOP_SET_ACL
5937 .set_acl = ll_set_acl,
5941 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
5943 const struct file_operations *fops = &ll_file_operations_noflock;
5945 if (test_bit(LL_SBI_FLOCK, sbi->ll_flags))
5946 fops = &ll_file_operations_flock;
5947 else if (test_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags))
5948 fops = &ll_file_operations;
5953 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5955 struct ll_inode_info *lli = ll_i2info(inode);
5956 struct cl_object *obj = lli->lli_clob;
5965 env = cl_env_get(&refcheck);
5967 RETURN(PTR_ERR(env));
5969 rc = cl_conf_set(env, lli->lli_clob, conf);
5973 if (conf->coc_opc == OBJECT_CONF_SET) {
5974 struct ldlm_lock *lock = conf->coc_lock;
5975 struct cl_layout cl = {
5979 LASSERT(lock != NULL);
5980 LASSERT(ldlm_has_layout(lock));
5982 /* it can only be allowed to match after layout is
5983 * applied to inode otherwise false layout would be
5984 * seen. Applying layout shoud happen before dropping
5985 * the intent lock. */
5986 ldlm_lock_allow_match(lock);
5988 rc = cl_object_layout_get(env, obj, &cl);
5993 DFID": layout version change: %u -> %u\n",
5994 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5996 ll_layout_version_set(lli, cl.cl_layout_gen);
6000 cl_env_put(env, &refcheck);
6002 RETURN(rc < 0 ? rc : 0);
6005 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
6006 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
6009 struct ll_sb_info *sbi = ll_i2sbi(inode);
6010 struct ptlrpc_request *req;
6017 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
6018 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
6019 lock->l_lvb_data, lock->l_lvb_len);
6021 if (lock->l_lvb_data != NULL)
6024 /* if layout lock was granted right away, the layout is returned
6025 * within DLM_LVB of dlm reply; otherwise if the lock was ever
6026 * blocked and then granted via completion ast, we have to fetch
6027 * layout here. Please note that we can't use the LVB buffer in
6028 * completion AST because it doesn't have a large enough buffer */
6029 rc = ll_get_default_mdsize(sbi, &lmmsize);
6033 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
6034 XATTR_NAME_LOV, lmmsize, &req);
6037 GOTO(out, rc = 0); /* empty layout */
6044 if (lmmsize == 0) /* empty layout */
6047 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
6049 GOTO(out, rc = -EFAULT);
6051 OBD_ALLOC_LARGE(lvbdata, lmmsize);
6052 if (lvbdata == NULL)
6053 GOTO(out, rc = -ENOMEM);
6055 memcpy(lvbdata, lmm, lmmsize);
6056 lock_res_and_lock(lock);
6057 if (unlikely(lock->l_lvb_data == NULL)) {
6058 lock->l_lvb_type = LVB_T_LAYOUT;
6059 lock->l_lvb_data = lvbdata;
6060 lock->l_lvb_len = lmmsize;
6063 unlock_res_and_lock(lock);
6066 OBD_FREE_LARGE(lvbdata, lmmsize);
6071 ptlrpc_req_finished(req);
6076 * Apply the layout to the inode. Layout lock is held and will be released
6079 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
6080 struct inode *inode)
6082 struct ll_inode_info *lli = ll_i2info(inode);
6083 struct ll_sb_info *sbi = ll_i2sbi(inode);
6084 struct ldlm_lock *lock;
6085 struct cl_object_conf conf;
6088 bool wait_layout = false;
6091 LASSERT(lustre_handle_is_used(lockh));
6093 lock = ldlm_handle2lock(lockh);
6094 LASSERT(lock != NULL);
6096 if (!ldlm_has_layout(lock))
6097 GOTO(out, rc = -EAGAIN);
6099 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
6100 PFID(&lli->lli_fid), inode);
6102 /* in case this is a caching lock and reinstate with new inode */
6103 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
6105 lock_res_and_lock(lock);
6106 lvb_ready = ldlm_is_lvb_ready(lock);
6107 unlock_res_and_lock(lock);
6109 /* checking lvb_ready is racy but this is okay. The worst case is
6110 * that multi processes may configure the file on the same time. */
6114 rc = ll_layout_fetch(inode, lock);
6118 /* for layout lock, lmm is stored in lock's lvb.
6119 * lvb_data is immutable if the lock is held so it's safe to access it
6122 * set layout to file. Unlikely this will fail as old layout was
6123 * surely eliminated */
6124 memset(&conf, 0, sizeof conf);
6125 conf.coc_opc = OBJECT_CONF_SET;
6126 conf.coc_inode = inode;
6127 conf.coc_lock = lock;
6128 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
6129 conf.u.coc_layout.lb_len = lock->l_lvb_len;
6130 rc = ll_layout_conf(inode, &conf);
6132 /* refresh layout failed, need to wait */
6133 wait_layout = rc == -EBUSY;
6136 LDLM_LOCK_PUT(lock);
6137 ldlm_lock_decref(lockh, mode);
6139 /* wait for IO to complete if it's still being used. */
6141 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
6142 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6144 memset(&conf, 0, sizeof conf);
6145 conf.coc_opc = OBJECT_CONF_WAIT;
6146 conf.coc_inode = inode;
6147 rc = ll_layout_conf(inode, &conf);
6151 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
6152 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
6158 * Issue layout intent RPC to MDS.
6159 * \param inode [in] file inode
6160 * \param intent [in] layout intent
6162 * \retval 0 on success
6163 * \retval < 0 error code
6165 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
6167 struct ll_inode_info *lli = ll_i2info(inode);
6168 struct ll_sb_info *sbi = ll_i2sbi(inode);
6169 struct md_op_data *op_data;
6170 struct lookup_intent it;
6171 struct ptlrpc_request *req;
6175 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
6176 0, 0, LUSTRE_OPC_ANY, NULL);
6177 if (IS_ERR(op_data))
6178 RETURN(PTR_ERR(op_data));
6180 op_data->op_data = intent;
6181 op_data->op_data_size = sizeof(*intent);
6183 memset(&it, 0, sizeof(it));
6184 it.it_op = IT_LAYOUT;
6185 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
6186 intent->li_opc == LAYOUT_INTENT_TRUNC)
6187 it.it_flags = FMODE_WRITE;
6189 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
6190 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6192 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
6193 &ll_md_blocking_ast, 0);
6194 if (it.it_request != NULL)
6195 ptlrpc_req_finished(it.it_request);
6196 it.it_request = NULL;
6198 ll_finish_md_op_data(op_data);
6200 /* set lock data in case this is a new lock */
6202 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
6204 ll_intent_drop_lock(&it);
6210 * This function checks if there exists a LAYOUT lock on the client side,
6211 * or enqueues it if it doesn't have one in cache.
6213 * This function will not hold layout lock so it may be revoked any time after
6214 * this function returns. Any operations depend on layout should be redone
6217 * This function should be called before lov_io_init() to get an uptodate
6218 * layout version, the caller should save the version number and after IO
6219 * is finished, this function should be called again to verify that layout
6220 * is not changed during IO time.
6222 int ll_layout_refresh(struct inode *inode, __u32 *gen)
6224 struct ll_inode_info *lli = ll_i2info(inode);
6225 struct ll_sb_info *sbi = ll_i2sbi(inode);
6226 struct lustre_handle lockh;
6227 struct layout_intent intent = {
6228 .li_opc = LAYOUT_INTENT_ACCESS,
6230 enum ldlm_mode mode;
6234 *gen = ll_layout_version_get(lli);
6235 if (!test_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags) ||
6236 *gen != CL_LAYOUT_GEN_NONE)
6240 LASSERT(fid_is_sane(ll_inode2fid(inode)));
6241 LASSERT(S_ISREG(inode->i_mode));
6243 /* take layout lock mutex to enqueue layout lock exclusively. */
6244 mutex_lock(&lli->lli_layout_mutex);
6247 /* mostly layout lock is caching on the local side, so try to
6248 * match it before grabbing layout lock mutex. */
6249 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
6250 LCK_CR | LCK_CW | LCK_PR |
6252 if (mode != 0) { /* hit cached lock */
6253 rc = ll_layout_lock_set(&lockh, mode, inode);
6259 rc = ll_layout_intent(inode, &intent);
6265 *gen = ll_layout_version_get(lli);
6266 mutex_unlock(&lli->lli_layout_mutex);
6272 * Issue layout intent RPC indicating where in a file an IO is about to write.
6274 * \param[in] inode file inode.
6275 * \param[in] ext write range with start offset of fille in bytes where
6276 * an IO is about to write, and exclusive end offset in
6279 * \retval 0 on success
6280 * \retval < 0 error code
6282 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
6283 struct lu_extent *ext)
6285 struct layout_intent intent = {
6287 .li_extent.e_start = ext->e_start,
6288 .li_extent.e_end = ext->e_end,
6293 rc = ll_layout_intent(inode, &intent);
6299 * This function send a restore request to the MDT
6301 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6303 struct hsm_user_request *hur;
6307 len = sizeof(struct hsm_user_request) +
6308 sizeof(struct hsm_user_item);
6309 OBD_ALLOC(hur, len);
6313 hur->hur_request.hr_action = HUA_RESTORE;
6314 hur->hur_request.hr_archive_id = 0;
6315 hur->hur_request.hr_flags = 0;
6316 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6317 sizeof(hur->hur_user_item[0].hui_fid));
6318 hur->hur_user_item[0].hui_extent.offset = offset;
6319 hur->hur_user_item[0].hui_extent.length = length;
6320 hur->hur_request.hr_itemcount = 1;
6321 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,