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 kunmap_atomic(kaddr);
463 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
464 if (!llcrypt_has_encryption_key(inode)) {
465 CDEBUG(D_SEC, "no enc key for "DFID"\n",
466 PFID(ll_inode2fid(inode)));
469 unsigned int offs = 0;
471 while (offs < PAGE_SIZE) {
472 /* decrypt only if page is not empty */
473 if (memcmp(page_address(page) + offs,
474 page_address(ZERO_PAGE(0)),
475 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
478 rc = llcrypt_decrypt_pagecache_blocks(page,
479 LUSTRE_ENCRYPTION_UNIT_SIZE,
484 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
489 flush_dcache_page(page);
490 SetPageUptodate(page);
497 #ifdef HAVE_READ_CACHE_PAGE_WANTS_FILE
498 static inline int ll_dom_read_folio(struct file *file, struct folio *folio0)
500 return ll_dom_readpage(file->private_data, folio_page(folio0, 0));
503 #define ll_dom_read_folio ll_dom_readpage
506 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
510 struct ll_inode_info *lli = ll_i2info(inode);
511 struct cl_object *obj = lli->lli_clob;
512 struct address_space *mapping = inode->i_mapping;
514 struct niobuf_remote *rnb;
515 struct mdt_body *body;
517 unsigned long index, start;
518 struct niobuf_local lnb;
527 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
531 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
532 if (rnb == NULL || rnb->rnb_len == 0)
535 /* LU-11595: Server may return whole file and that is OK always or
536 * it may return just file tail and its offset must be aligned with
537 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
538 * smaller then offset may be not aligned and that data is just ignored.
540 if (rnb->rnb_offset & ~PAGE_MASK)
543 /* Server returns whole file or just file tail if it fills in reply
544 * buffer, in both cases total size should be equal to the file size.
546 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
547 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
548 !(inode && IS_ENCRYPTED(inode))) {
549 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
550 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
551 rnb->rnb_len, body->mbo_dom_size);
555 env = cl_env_get(&refcheck);
558 io = vvp_env_thread_io(env);
560 io->ci_ignore_layout = 1;
561 rc = cl_io_init(env, io, CIT_MISC, obj);
565 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
566 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
568 data = (char *)rnb + sizeof(*rnb);
570 lnb.lnb_file_offset = rnb->rnb_offset;
571 start = lnb.lnb_file_offset >> PAGE_SHIFT;
573 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
574 lnb.lnb_page_offset = 0;
576 struct cl_page *page;
578 lnb.lnb_data = data + (index << PAGE_SHIFT);
579 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
580 if (lnb.lnb_len > PAGE_SIZE)
581 lnb.lnb_len = PAGE_SIZE;
583 vmpage = ll_read_cache_page(mapping, index + start,
584 ll_dom_read_folio, &lnb);
585 if (IS_ERR(vmpage)) {
586 CWARN("%s: cannot fill page %lu for "DFID
587 " with data: rc = %li\n",
588 ll_i2sbi(inode)->ll_fsname, index + start,
589 PFID(lu_object_fid(&obj->co_lu)),
594 if (vmpage->mapping == NULL) {
597 /* page was truncated */
600 /* attach VM page to CL page cache */
601 page = cl_page_find(env, obj, vmpage->index, vmpage,
604 ClearPageUptodate(vmpage);
609 SetPageUptodate(vmpage);
610 cl_page_put(env, page);
614 } while (rnb->rnb_len > (index << PAGE_SHIFT));
618 cl_env_put(env, &refcheck);
623 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
624 struct lookup_intent *itp)
626 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
627 struct dentry *parent = de->d_parent;
630 struct md_op_data *op_data;
631 struct ptlrpc_request *req = NULL;
635 LASSERT(parent != NULL);
636 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
638 /* if server supports open-by-fid, or file name is invalid, don't pack
639 * name in open request */
640 if (CFS_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
641 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
643 len = de->d_name.len;
644 name = kmalloc(len + 1, GFP_NOFS);
649 spin_lock(&de->d_lock);
650 if (len != de->d_name.len) {
651 spin_unlock(&de->d_lock);
655 memcpy(name, de->d_name.name, len);
657 spin_unlock(&de->d_lock);
659 if (!lu_name_is_valid_2(name, len)) {
665 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
666 name, len, 0, LUSTRE_OPC_OPEN, NULL);
667 if (IS_ERR(op_data)) {
669 RETURN(PTR_ERR(op_data));
671 op_data->op_data = lmm;
672 op_data->op_data_size = lmmsize;
674 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_OPEN_DELAY, cfs_fail_val);
676 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
677 &ll_md_blocking_ast, 0);
679 ll_finish_md_op_data(op_data);
681 /* reason for keep own exit path - don`t flood log
682 * with messages with -ESTALE errors.
684 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
685 it_open_error(DISP_OPEN_OPEN, itp))
687 ll_release_openhandle(de, itp);
691 if (it_disposition(itp, DISP_LOOKUP_NEG))
692 GOTO(out, rc = -ENOENT);
694 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
695 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
696 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
700 rc = ll_prep_inode(&de->d_inode, &req->rq_pill, NULL, itp);
702 if (!rc && itp->it_lock_mode) {
705 /* If we got a lock back and it has a LOOKUP bit set,
706 * make sure the dentry is marked as valid so we can find it.
707 * We don't need to care about actual hashing since other bits
708 * of kernel will deal with that later.
710 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
711 if (bits & MDS_INODELOCK_LOOKUP)
712 d_lustre_revalidate(de);
714 /* if DoM bit returned along with LAYOUT bit then there
715 * can be read-on-open data returned.
717 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
718 ll_dom_finish_open(de->d_inode, req);
720 /* open may not fetch LOOKUP lock, update dir depth and default LMV
723 if (!rc && S_ISDIR(de->d_inode->i_mode))
724 ll_update_dir_depth_dmv(parent->d_inode, de);
727 ptlrpc_req_finished(req);
728 ll_intent_drop_lock(itp);
730 /* We did open by fid, but by the time we got to the server, the object
731 * disappeared. This is possible if the object was unlinked, but it's
732 * also possible if the object was unlinked by a rename. In the case
733 * of an object renamed over our existing one, we can't fail this open.
734 * O_CREAT also goes through this path if we had an existing dentry,
735 * and it's obviously wrong to return ENOENT for O_CREAT.
737 * Instead let's return -ESTALE, and the VFS will retry the open with
738 * LOOKUP_REVAL, which we catch in ll_revalidate_dentry and fail to
739 * revalidate, causing a lookup. This causes extra lookups in the case
740 * where we had a dentry in cache but the file is being unlinked and we
741 * lose the race with unlink, but this should be very rare.
749 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
750 struct obd_client_handle *och)
752 struct mdt_body *body;
754 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
755 och->och_open_handle = body->mbo_open_handle;
756 och->och_fid = body->mbo_fid1;
757 och->och_lease_handle.cookie = it->it_lock_handle;
758 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
759 och->och_flags = it->it_flags;
761 return md_set_open_replay_data(md_exp, och, it);
764 static int ll_local_open(struct file *file, struct lookup_intent *it,
765 struct ll_file_data *fd, struct obd_client_handle *och)
767 struct inode *inode = file_inode(file);
770 LASSERT(!file->private_data);
777 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
782 file->private_data = fd;
783 ll_readahead_init(inode, &fd->fd_ras);
784 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
789 void ll_track_file_opens(struct inode *inode)
791 struct ll_inode_info *lli = ll_i2info(inode);
792 struct ll_sb_info *sbi = ll_i2sbi(inode);
794 /* do not skew results with delays from never-opened inodes */
795 if (ktime_to_ns(lli->lli_close_fd_time))
796 ll_stats_ops_tally(sbi, LPROC_LL_INODE_OPCLTM,
797 ktime_us_delta(ktime_get(), lli->lli_close_fd_time));
799 if (ktime_after(ktime_get(),
800 ktime_add_ms(lli->lli_close_fd_time,
801 sbi->ll_oc_max_ms))) {
802 lli->lli_open_fd_count = 1;
803 lli->lli_close_fd_time = ns_to_ktime(0);
805 lli->lli_open_fd_count++;
808 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_OCOUNT,
809 lli->lli_open_fd_count);
812 /* Open a file, and (for the very first open) create objects on the OSTs at
813 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
814 * creation or open until ll_lov_setstripe() ioctl is called.
816 * If we already have the stripe MD locally then we don't request it in
817 * md_open(), by passing a lmm_size = 0.
819 * It is up to the application to ensure no other processes open this file
820 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
821 * used. We might be able to avoid races of that sort by getting lli_open_sem
822 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
823 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
825 int ll_file_open(struct inode *inode, struct file *file)
827 struct ll_inode_info *lli = ll_i2info(inode);
828 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
829 .it_flags = file->f_flags };
830 struct obd_client_handle **och_p = NULL;
831 __u64 *och_usecount = NULL;
832 struct ll_file_data *fd;
833 ktime_t kstart = ktime_get();
837 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
838 PFID(ll_inode2fid(inode)), inode, file->f_flags);
840 it = file->private_data; /* XXX: compat macro */
841 file->private_data = NULL; /* prevent ll_local_open assertion */
843 if (S_ISREG(inode->i_mode)) {
844 rc = ll_file_open_encrypt(inode, file);
846 if (it && it->it_disposition)
847 ll_release_openhandle(file_dentry(file), it);
848 GOTO(out_nofiledata, rc);
852 fd = ll_file_data_get();
854 GOTO(out_nofiledata, rc = -ENOMEM);
857 if (S_ISDIR(inode->i_mode))
858 ll_authorize_statahead(inode, fd);
860 ll_track_file_opens(inode);
861 if (is_root_inode(inode)) {
862 file->private_data = fd;
866 if (!it || !it->it_disposition) {
867 /* Convert f_flags into access mode. We cannot use file->f_mode,
868 * because everything but O_ACCMODE mask was stripped from
870 if ((oit.it_flags + 1) & O_ACCMODE)
872 if (file->f_flags & O_TRUNC)
873 oit.it_flags |= FMODE_WRITE;
875 /* kernel only call f_op->open in dentry_open. filp_open calls
876 * dentry_open after call to open_namei that checks permissions.
877 * Only nfsd_open call dentry_open directly without checking
878 * permissions and because of that this code below is safe.
880 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
881 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
883 /* We do not want O_EXCL here, presumably we opened the file
884 * already? XXX - NFS implications? */
885 oit.it_flags &= ~O_EXCL;
887 /* bug20584, if "it_flags" contains O_CREAT, the file will be
888 * created if necessary, then "IT_CREAT" should be set to keep
889 * consistent with it */
890 if (oit.it_flags & O_CREAT)
891 oit.it_op |= IT_CREAT;
897 /* Let's see if we have file open on MDS already. */
898 if (it->it_flags & FMODE_WRITE) {
899 och_p = &lli->lli_mds_write_och;
900 och_usecount = &lli->lli_open_fd_write_count;
901 } else if (it->it_flags & FMODE_EXEC) {
902 och_p = &lli->lli_mds_exec_och;
903 och_usecount = &lli->lli_open_fd_exec_count;
905 och_p = &lli->lli_mds_read_och;
906 och_usecount = &lli->lli_open_fd_read_count;
909 mutex_lock(&lli->lli_och_mutex);
910 if (*och_p) { /* Open handle is present */
911 if (it_disposition(it, DISP_OPEN_OPEN)) {
912 /* Well, there's extra open request that we do not need,
913 * let's close it somehow. This will decref request. */
914 rc = it_open_error(DISP_OPEN_OPEN, it);
916 mutex_unlock(&lli->lli_och_mutex);
917 GOTO(out_openerr, rc);
920 ll_release_openhandle(file_dentry(file), it);
924 rc = ll_local_open(file, it, fd, NULL);
927 mutex_unlock(&lli->lli_och_mutex);
928 GOTO(out_openerr, rc);
931 LASSERT(*och_usecount == 0);
932 if (!it->it_disposition) {
933 struct dentry *dentry = file_dentry(file);
934 struct ll_sb_info *sbi = ll_i2sbi(inode);
935 int open_threshold = sbi->ll_oc_thrsh_count;
937 /* We cannot just request lock handle now, new ELC code
938 * means that one of other OPEN locks for this file
939 * could be cancelled, and since blocking ast handler
940 * would attempt to grab och_mutex as well, that would
941 * result in a deadlock
943 mutex_unlock(&lli->lli_och_mutex);
945 * Normally called under two situations:
946 * 1. fhandle / NFS export.
947 * 2. A race/condition on MDS resulting in no open
948 * handle to be returned from LOOKUP|OPEN request,
949 * for example if the target entry was a symlink.
951 * For NFSv3 we need to always cache the open lock
952 * for pre 5.5 Linux kernels.
954 * After reaching number of opens of this inode
955 * we always ask for an open lock on it to handle
956 * bad userspace actors that open and close files
957 * in a loop for absolutely no good reason
959 /* fhandle / NFS path. */
960 if (lli->lli_open_thrsh_count != UINT_MAX)
961 open_threshold = lli->lli_open_thrsh_count;
963 if (filename_is_volatile(dentry->d_name.name,
966 /* There really is nothing here, but this
967 * make this more readable I think.
968 * We do not want openlock for volatile
969 * files under any circumstances
971 } else if (open_threshold > 0) {
972 /* Take MDS_OPEN_LOCK with many opens */
973 if (lli->lli_open_fd_count >= open_threshold)
974 it->it_flags |= MDS_OPEN_LOCK;
976 /* If this is open after we just closed */
977 else if (ktime_before(ktime_get(),
978 ktime_add_ms(lli->lli_close_fd_time,
979 sbi->ll_oc_thrsh_ms)))
980 it->it_flags |= MDS_OPEN_LOCK;
984 * Always specify MDS_OPEN_BY_FID because we don't want
985 * to get file with different fid.
987 it->it_flags |= MDS_OPEN_BY_FID;
988 rc = ll_intent_file_open(dentry, NULL, 0, it);
990 GOTO(out_openerr, rc);
994 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
996 GOTO(out_och_free, rc = -ENOMEM);
1000 /* md_intent_lock() didn't get a request ref if there was an
1001 * open error, so don't do cleanup on the request here
1003 /* XXX (green): Should not we bail out on any error here, not
1004 * just open error? */
1005 rc = it_open_error(DISP_OPEN_OPEN, it);
1007 GOTO(out_och_free, rc);
1009 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
1010 "inode %p: disposition %x, status %d\n", inode,
1011 it_disposition(it, ~0), it->it_status);
1013 rc = ll_local_open(file, it, fd, *och_p);
1015 GOTO(out_och_free, rc);
1018 rc = pcc_file_open(inode, file);
1020 GOTO(out_och_free, rc);
1022 mutex_unlock(&lli->lli_och_mutex);
1026 /* Must do this outside lli_och_mutex lock to prevent deadlock where
1027 different kind of OPEN lock for this same inode gets cancelled
1028 by ldlm_cancel_lru */
1029 if (!S_ISREG(inode->i_mode))
1030 GOTO(out_och_free, rc);
1031 cl_lov_delay_create_clear(&file->f_flags);
1032 GOTO(out_och_free, rc);
1036 if (och_p && *och_p) {
1037 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
1038 *och_p = NULL; /* OBD_FREE writes some magic there */
1041 mutex_unlock(&lli->lli_och_mutex);
1044 if (lli->lli_opendir_key == fd)
1045 ll_deauthorize_statahead(inode, fd);
1048 ll_file_data_put(fd);
1050 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
1051 ktime_us_delta(ktime_get(), kstart));
1055 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
1056 ptlrpc_req_finished(it->it_request);
1057 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1063 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
1064 struct ldlm_lock_desc *desc, void *data, int flag)
1067 struct lustre_handle lockh;
1071 case LDLM_CB_BLOCKING:
1072 ldlm_lock2handle(lock, &lockh);
1073 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
1075 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1079 case LDLM_CB_CANCELING:
1087 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1088 * and save it as fd->fd_och so as to force client to reopen the file even
1089 * if it has an open lock in cache already.
1091 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1092 struct lustre_handle *old_open_handle)
1094 struct ll_inode_info *lli = ll_i2info(inode);
1095 struct ll_file_data *fd = file->private_data;
1096 struct obd_client_handle **och_p;
1097 __u64 *och_usecount;
1101 /* Get the openhandle of the file */
1102 mutex_lock(&lli->lli_och_mutex);
1103 if (fd->fd_lease_och != NULL)
1104 GOTO(out_unlock, rc = -EBUSY);
1106 if (fd->fd_och == NULL) {
1107 if (file->f_mode & FMODE_WRITE) {
1108 LASSERT(lli->lli_mds_write_och != NULL);
1109 och_p = &lli->lli_mds_write_och;
1110 och_usecount = &lli->lli_open_fd_write_count;
1112 LASSERT(lli->lli_mds_read_och != NULL);
1113 och_p = &lli->lli_mds_read_och;
1114 och_usecount = &lli->lli_open_fd_read_count;
1117 if (*och_usecount > 1)
1118 GOTO(out_unlock, rc = -EBUSY);
1120 fd->fd_och = *och_p;
1125 *old_open_handle = fd->fd_och->och_open_handle;
1129 mutex_unlock(&lli->lli_och_mutex);
1134 * Release ownership on lli_mds_*_och when putting back a file lease.
1136 static int ll_lease_och_release(struct inode *inode, struct file *file)
1138 struct ll_inode_info *lli = ll_i2info(inode);
1139 struct ll_file_data *fd = file->private_data;
1140 struct obd_client_handle **och_p;
1141 struct obd_client_handle *old_och = NULL;
1142 __u64 *och_usecount;
1146 mutex_lock(&lli->lli_och_mutex);
1147 if (file->f_mode & FMODE_WRITE) {
1148 och_p = &lli->lli_mds_write_och;
1149 och_usecount = &lli->lli_open_fd_write_count;
1151 och_p = &lli->lli_mds_read_och;
1152 och_usecount = &lli->lli_open_fd_read_count;
1155 /* The file may have been open by another process (broken lease) so
1156 * *och_p is not NULL. In this case we should simply increase usecount
1159 if (*och_p != NULL) {
1160 old_och = fd->fd_och;
1163 *och_p = fd->fd_och;
1167 mutex_unlock(&lli->lli_och_mutex);
1169 if (old_och != NULL)
1170 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1176 * Acquire a lease and open the file.
1178 static struct obd_client_handle *
1179 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1182 struct lookup_intent it = { .it_op = IT_OPEN };
1183 struct ll_sb_info *sbi = ll_i2sbi(inode);
1184 struct md_op_data *op_data;
1185 struct ptlrpc_request *req = NULL;
1186 struct lustre_handle old_open_handle = { 0 };
1187 struct obd_client_handle *och = NULL;
1192 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1193 RETURN(ERR_PTR(-EINVAL));
1196 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1197 RETURN(ERR_PTR(-EPERM));
1199 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1201 RETURN(ERR_PTR(rc));
1206 RETURN(ERR_PTR(-ENOMEM));
1208 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1209 LUSTRE_OPC_ANY, NULL);
1210 if (IS_ERR(op_data))
1211 GOTO(out, rc = PTR_ERR(op_data));
1213 /* To tell the MDT this openhandle is from the same owner */
1214 op_data->op_open_handle = old_open_handle;
1216 it.it_flags = fmode | open_flags;
1217 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1218 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1219 &ll_md_blocking_lease_ast,
1220 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1221 * it can be cancelled which may mislead applications that the lease is
1223 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1224 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1225 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1226 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1227 ll_finish_md_op_data(op_data);
1228 ptlrpc_req_finished(req);
1230 GOTO(out_release_it, rc);
1232 if (it_disposition(&it, DISP_LOOKUP_NEG))
1233 GOTO(out_release_it, rc = -ENOENT);
1235 rc = it_open_error(DISP_OPEN_OPEN, &it);
1237 GOTO(out_release_it, rc);
1239 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1240 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1242 GOTO(out_release_it, rc);
1244 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1245 GOTO(out_close, rc = -EOPNOTSUPP);
1247 /* already get lease, handle lease lock */
1248 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1249 if (!it.it_lock_mode ||
1250 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1251 /* open lock must return for lease */
1252 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1253 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1255 GOTO(out_close, rc = -EPROTO);
1258 ll_intent_release(&it);
1262 /* Cancel open lock */
1263 if (it.it_lock_mode != 0) {
1264 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1266 it.it_lock_mode = 0;
1267 och->och_lease_handle.cookie = 0ULL;
1269 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1271 CERROR("%s: error closing file "DFID": %d\n",
1272 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1273 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1275 ll_intent_release(&it);
1279 RETURN(ERR_PTR(rc));
1283 * Check whether a layout swap can be done between two inodes.
1285 * \param[in] inode1 First inode to check
1286 * \param[in] inode2 Second inode to check
1288 * \retval 0 on success, layout swap can be performed between both inodes
1289 * \retval negative error code if requirements are not met
1291 static int ll_check_swap_layouts_validity(struct inode *inode1,
1292 struct inode *inode2)
1294 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1297 if (inode_permission(&init_user_ns, inode1, MAY_WRITE) ||
1298 inode_permission(&init_user_ns, inode2, MAY_WRITE))
1301 if (inode1->i_sb != inode2->i_sb)
1307 static int ll_swap_layouts_close(struct obd_client_handle *och,
1308 struct inode *inode, struct inode *inode2)
1310 const struct lu_fid *fid1 = ll_inode2fid(inode);
1311 const struct lu_fid *fid2;
1315 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1316 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1318 rc = ll_check_swap_layouts_validity(inode, inode2);
1320 GOTO(out_free_och, rc);
1322 /* We now know that inode2 is a lustre inode */
1323 fid2 = ll_inode2fid(inode2);
1325 rc = lu_fid_cmp(fid1, fid2);
1327 GOTO(out_free_och, rc = -EINVAL);
1329 /* Close the file and {swap,merge} layouts between inode & inode2.
1330 * NB: local lease handle is released in mdc_close_intent_pack()
1331 * because we still need it to pack l_remote_handle to MDT. */
1332 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1335 och = NULL; /* freed in ll_close_inode_openhandle() */
1345 * Release lease and close the file.
1346 * It will check if the lease has ever broken.
1348 static int ll_lease_close_intent(struct obd_client_handle *och,
1349 struct inode *inode,
1350 bool *lease_broken, enum mds_op_bias bias,
1353 struct ldlm_lock *lock;
1354 bool cancelled = true;
1358 lock = ldlm_handle2lock(&och->och_lease_handle);
1360 lock_res_and_lock(lock);
1361 cancelled = ldlm_is_cancel(lock);
1362 unlock_res_and_lock(lock);
1363 LDLM_LOCK_PUT(lock);
1366 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1367 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1369 if (lease_broken != NULL)
1370 *lease_broken = cancelled;
1372 if (!cancelled && !bias)
1373 ldlm_cli_cancel(&och->och_lease_handle, 0);
1375 if (cancelled) { /* no need to excute intent */
1380 rc = ll_close_inode_openhandle(inode, och, bias, data);
1384 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1387 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1391 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1393 static int ll_lease_file_resync(struct obd_client_handle *och,
1394 struct inode *inode, void __user *uarg)
1396 struct ll_sb_info *sbi = ll_i2sbi(inode);
1397 struct md_op_data *op_data;
1398 struct ll_ioc_lease_id ioc;
1399 __u64 data_version_unused;
1403 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1404 LUSTRE_OPC_ANY, NULL);
1405 if (IS_ERR(op_data))
1406 RETURN(PTR_ERR(op_data));
1408 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
1411 /* before starting file resync, it's necessary to clean up page cache
1412 * in client memory, otherwise once the layout version is increased,
1413 * writing back cached data will be denied the OSTs. */
1414 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1418 op_data->op_lease_handle = och->och_lease_handle;
1419 op_data->op_mirror_id = ioc.lil_mirror_id;
1420 rc = md_file_resync(sbi->ll_md_exp, op_data);
1426 ll_finish_md_op_data(op_data);
1430 static int ll_merge_attr_nolock(const struct lu_env *env, struct inode *inode)
1432 struct ll_inode_info *lli = ll_i2info(inode);
1433 struct cl_object *obj = lli->lli_clob;
1434 struct cl_attr *attr = vvp_env_thread_attr(env);
1442 /* Merge timestamps the most recently obtained from MDS with
1443 * timestamps obtained from OSTs.
1445 * Do not overwrite atime of inode because it may be refreshed
1446 * by file_accessed() function. If the read was served by cache
1447 * data, there is no RPC to be sent so that atime may not be
1448 * transferred to OSTs at all. MDT only updates atime at close time
1449 * if it's at least 'mdd.*.atime_diff' older.
1450 * All in all, the atime in Lustre does not strictly comply with
1451 * POSIX. Solving this problem needs to send an RPC to MDT for each
1452 * read, this will hurt performance.
1454 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1455 inode->i_atime.tv_sec < lli->lli_atime)
1456 inode->i_atime.tv_sec = lli->lli_atime;
1458 inode->i_mtime.tv_sec = lli->lli_mtime;
1459 inode->i_ctime.tv_sec = lli->lli_ctime;
1461 mtime = inode->i_mtime.tv_sec;
1462 atime = inode->i_atime.tv_sec;
1463 ctime = inode->i_ctime.tv_sec;
1465 cl_object_attr_lock(obj);
1466 if (CFS_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1469 rc = cl_object_attr_get(env, obj, attr);
1470 cl_object_attr_unlock(obj);
1473 GOTO(out, rc = (rc == -ENODATA ? 0 : rc));
1475 if (atime < attr->cat_atime)
1476 atime = attr->cat_atime;
1478 if (ctime < attr->cat_ctime)
1479 ctime = attr->cat_ctime;
1481 if (mtime < attr->cat_mtime)
1482 mtime = attr->cat_mtime;
1484 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu i_blocks %llu\n",
1485 PFID(&lli->lli_fid), attr->cat_size, attr->cat_blocks);
1487 if (llcrypt_require_key(inode) == -ENOKEY) {
1488 /* Without the key, round up encrypted file size to next
1489 * LUSTRE_ENCRYPTION_UNIT_SIZE. Clear text size is put in
1490 * lli_lazysize for proper file size setting at close time.
1492 lli->lli_attr_valid |= OBD_MD_FLLAZYSIZE;
1493 lli->lli_lazysize = attr->cat_size;
1494 attr->cat_size = round_up(attr->cat_size,
1495 LUSTRE_ENCRYPTION_UNIT_SIZE);
1497 i_size_write(inode, attr->cat_size);
1498 inode->i_blocks = attr->cat_blocks;
1500 inode->i_mtime.tv_sec = mtime;
1501 inode->i_atime.tv_sec = atime;
1502 inode->i_ctime.tv_sec = ctime;
1509 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1513 ll_inode_size_lock(inode);
1514 rc = ll_merge_attr_nolock(env, inode);
1515 ll_inode_size_unlock(inode);
1520 /* Use to update size and blocks on inode for LSOM if there is no contention */
1521 int ll_merge_attr_try(const struct lu_env *env, struct inode *inode)
1525 if (ll_inode_size_trylock(inode)) {
1526 rc = ll_merge_attr_nolock(env, inode);
1527 ll_inode_size_unlock(inode);
1534 * Set designated mirror for I/O.
1536 * So far only read, write, and truncated can support to issue I/O to
1537 * designated mirror.
1539 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1541 struct ll_file_data *fd = file->private_data;
1543 /* clear layout version for generic(non-resync) I/O in case it carries
1544 * stale layout version due to I/O restart */
1545 io->ci_layout_version = 0;
1547 /* FLR: disable non-delay for designated mirror I/O because obviously
1548 * only one mirror is available */
1549 if (fd->fd_designated_mirror > 0) {
1551 io->ci_designated_mirror = fd->fd_designated_mirror;
1552 io->ci_layout_version = fd->fd_layout_version;
1555 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1556 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1560 * This is relatime_need_update() from Linux 5.17, which is not exported.
1562 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1563 struct timespec64 now)
1566 if (!(mnt->mnt_flags & MNT_RELATIME))
1569 * Is mtime younger than atime? If yes, update atime:
1571 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1574 * Is ctime younger than atime? If yes, update atime:
1576 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1580 * Is the previous atime value older than a day? If yes,
1583 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1586 * Good, we can skip the atime update:
1592 * Very similar to kernel function: !__atime_needs_update()
1594 static bool file_is_noatime(const struct file *file)
1596 struct vfsmount *mnt = file->f_path.mnt;
1597 struct inode *inode = file_inode((struct file *)file);
1598 struct timespec64 now;
1600 if (file->f_flags & O_NOATIME)
1603 if (inode->i_flags & S_NOATIME)
1606 if (IS_NOATIME(inode))
1609 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1612 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1615 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1618 now = current_time(inode);
1620 if (!relatime_need_update(mnt, inode, now))
1626 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1627 struct vvp_io_args *args)
1629 struct inode *inode = file_inode(file);
1630 struct ll_file_data *fd = file->private_data;
1632 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1633 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1635 if (iot == CIT_WRITE) {
1636 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1637 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1638 file->f_flags & O_DIRECT ||
1640 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1641 io->u.ci_wr.wr_sync |= !!(args &&
1642 (args->u.normal.via_iocb->ki_flags &
1648 io->ci_iocb_nowait = !!(args &&
1649 (args->u.normal.via_iocb->ki_flags &
1653 io->ci_obj = ll_i2info(inode)->lli_clob;
1654 io->ci_lockreq = CILR_MAYBE;
1655 if (ll_file_nolock(file)) {
1656 io->ci_lockreq = CILR_NEVER;
1657 io->ci_no_srvlock = 1;
1658 } else if (file->f_flags & O_APPEND) {
1659 io->ci_lockreq = CILR_MANDATORY;
1661 io->ci_noatime = file_is_noatime(file);
1662 io->ci_async_readahead = false;
1664 /* FLR: only use non-delay I/O for read as there is only one
1665 * avaliable mirror for write. */
1666 io->ci_ndelay = !(iot == CIT_WRITE);
1668 ll_io_set_mirror(io, file);
1671 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1674 struct ll_inode_info *lli = ll_i2info(inode);
1675 struct ll_sb_info *sbi = ll_i2sbi(inode);
1676 enum obd_heat_type sample_type;
1677 enum obd_heat_type iobyte_type;
1678 __u64 now = ktime_get_real_seconds();
1680 if (!ll_sbi_has_file_heat(sbi) ||
1681 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1684 if (iot == CIT_READ) {
1685 sample_type = OBD_HEAT_READSAMPLE;
1686 iobyte_type = OBD_HEAT_READBYTE;
1687 } else if (iot == CIT_WRITE) {
1688 sample_type = OBD_HEAT_WRITESAMPLE;
1689 iobyte_type = OBD_HEAT_WRITEBYTE;
1694 spin_lock(&lli->lli_heat_lock);
1695 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1696 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1697 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1698 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1699 spin_unlock(&lli->lli_heat_lock);
1703 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1704 struct file *file, enum cl_io_type iot,
1705 loff_t *ppos, size_t count)
1707 struct vvp_io *vio = vvp_env_io(env);
1708 struct inode *inode = file_inode(file);
1709 struct ll_inode_info *lli = ll_i2info(inode);
1710 struct ll_sb_info *sbi = ll_i2sbi(inode);
1711 struct ll_file_data *fd = file->private_data;
1712 struct range_lock range;
1713 bool range_locked = false;
1719 unsigned int retried = 0, dio_lock = 0;
1720 bool is_aio = false;
1721 bool is_parallel_dio = false;
1722 struct cl_dio_aio *ci_dio_aio = NULL;
1724 bool partial_io = false;
1725 size_t max_io_pages, max_cached_pages;
1729 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1730 file_dentry(file)->d_name.name,
1731 iot == CIT_READ ? "read" : "write", *ppos, count);
1733 max_io_pages = PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT;
1734 max_cached_pages = sbi->ll_cache->ccc_lru_max;
1735 if (max_io_pages > (max_cached_pages >> 2))
1736 max_io_pages = max_cached_pages >> 2;
1738 io = vvp_env_thread_io(env);
1739 if (file->f_flags & O_DIRECT) {
1740 if (file->f_flags & O_APPEND)
1742 if (!is_sync_kiocb(args->u.normal.via_iocb))
1745 /* the kernel does not support AIO on pipes, and parallel DIO
1746 * uses part of the AIO path, so we must not do parallel dio
1749 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1752 if (!ll_sbi_has_parallel_dio(sbi))
1753 is_parallel_dio = false;
1755 ci_dio_aio = cl_dio_aio_alloc(args->u.normal.via_iocb,
1756 ll_i2info(inode)->lli_clob, is_aio);
1758 GOTO(out, rc = -ENOMEM);
1763 * IO block size need be aware of cached page limit, otherwise
1764 * if we have small max_cached_mb but large block IO issued, io
1765 * could not be finished and blocked whole client.
1767 if (file->f_flags & O_DIRECT)
1770 per_bytes = min(max_io_pages << PAGE_SHIFT, count);
1771 partial_io = per_bytes < count;
1772 io = vvp_env_thread_io(env);
1773 ll_io_init(io, file, iot, args);
1774 io->ci_dio_aio = ci_dio_aio;
1775 io->ci_dio_lock = dio_lock;
1776 io->ci_ndelay_tried = retried;
1777 io->ci_parallel_dio = is_parallel_dio;
1779 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1780 if (file->f_flags & O_APPEND)
1781 range_lock_init(&range, 0, LUSTRE_EOF);
1783 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1785 vio->vui_fd = file->private_data;
1786 vio->vui_iter = args->u.normal.via_iter;
1787 vio->vui_iocb = args->u.normal.via_iocb;
1788 /* Direct IO reads must also take range lock,
1789 * or multiple reads will try to work on the same pages
1790 * See LU-6227 for details.
1792 if (((iot == CIT_WRITE) ||
1793 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1794 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1795 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1797 rc = range_lock(&lli->lli_write_tree, &range);
1801 range_locked = true;
1804 ll_cl_add(inode, env, io, LCC_RW);
1805 rc = cl_io_loop(env, io);
1806 ll_cl_remove(inode, env);
1808 /* cl_io_rw_init() handled IO */
1812 if (io->ci_dio_aio && !is_aio) {
1813 struct cl_sync_io *anchor = &io->ci_dio_aio->cda_sync;
1815 /* for dio, EIOCBQUEUED is an implementation detail,
1816 * and we don't return it to userspace
1818 if (rc == -EIOCBQUEUED)
1821 /* N/B: parallel DIO may be disabled during i/o submission;
1822 * if that occurs, I/O shifts to sync, so it's all resolved
1823 * before we get here, and this wait call completes
1826 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1832 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1834 range_unlock(&lli->lli_write_tree, &range);
1835 range_locked = false;
1838 if (io->ci_nob > 0) {
1840 result += io->ci_nob;
1841 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1845 count -= io->ci_nob;
1847 /* prepare IO restart */
1849 args->u.normal.via_iter = vio->vui_iter;
1853 * Reexpand iov count because it was zero
1856 iov_iter_reexpand(vio->vui_iter, count);
1857 if (per_bytes == io->ci_nob)
1858 io->ci_need_restart = 1;
1862 cl_io_fini(env, io);
1865 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1866 file->f_path.dentry->d_name.name,
1867 iot, rc, result, io->ci_need_restart);
1869 if ((!rc || rc == -ENODATA || rc == -ENOLCK || rc == -EIOCBQUEUED) &&
1870 count > 0 && io->ci_need_restart && retries-- > 0) {
1872 "%s: restart %s from ppos=%lld count=%zu retries=%u ret=%zd: rc = %d\n",
1873 file_dentry(file)->d_name.name,
1874 iot == CIT_READ ? "read" : "write",
1875 *ppos, count, retries, result, rc);
1876 /* preserve the tried count for FLR */
1877 retried = io->ci_ndelay_tried;
1878 dio_lock = io->ci_dio_lock;
1882 if (io->ci_dio_aio) {
1883 /* set the number of bytes successfully moved in the aio */
1885 io->ci_dio_aio->cda_bytes = result;
1887 * VFS will call aio_complete() if no -EIOCBQUEUED
1888 * is returned for AIO, so we can not call aio_complete()
1889 * in our end_io(). (cda_no_aio_complete is always set for
1892 * NB: Setting cda_no_aio_complete like this is safe because
1893 * the atomic_dec_and_lock in cl_sync_io_note has implicit
1894 * memory barriers, so this will be seen by whichever thread
1895 * completes the DIO/AIO, even if it's not this one.
1897 if (is_aio && rc != -EIOCBQUEUED)
1898 io->ci_dio_aio->cda_no_aio_complete = 1;
1899 /* if an aio enqueued successfully (-EIOCBQUEUED), then Lustre
1900 * will call aio_complete rather than the vfs, so we return 0
1901 * to tell the VFS we're handling it
1903 else if (is_aio) /* rc == -EIOCBQUEUED */
1906 * Drop the reference held by the llite layer on this top level
1909 * For DIO, this frees it here, since IO is complete, and for
1910 * AIO, we will call aio_complete() (and then free this top
1911 * level context) once all the outstanding chunks of this AIO
1914 cl_sync_io_note(env, &io->ci_dio_aio->cda_sync,
1915 rc == -EIOCBQUEUED ? 0 : rc);
1917 LASSERT(io->ci_dio_aio->cda_creator_free);
1918 cl_dio_aio_free(env, io->ci_dio_aio);
1919 io->ci_dio_aio = NULL;
1923 if (iot == CIT_READ) {
1925 ll_stats_ops_tally(ll_i2sbi(inode),
1926 LPROC_LL_READ_BYTES, result);
1927 } else if (iot == CIT_WRITE) {
1929 ll_stats_ops_tally(ll_i2sbi(inode),
1930 LPROC_LL_WRITE_BYTES, result);
1931 fd->fd_write_failed = false;
1932 } else if (result == 0 && rc == 0) {
1935 fd->fd_write_failed = true;
1937 fd->fd_write_failed = false;
1938 } else if (rc != -ERESTARTSYS) {
1939 fd->fd_write_failed = true;
1943 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1945 ll_heat_add(inode, iot, result);
1947 RETURN(result > 0 ? result : rc);
1951 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1952 * especially for small I/O.
1954 * To serve a read request, CLIO has to create and initialize a cl_io and
1955 * then request DLM lock. This has turned out to have siginificant overhead
1956 * and affects the performance of small I/O dramatically.
1958 * It's not necessary to create a cl_io for each I/O. Under the help of read
1959 * ahead, most of the pages being read are already in memory cache and we can
1960 * read those pages directly because if the pages exist, the corresponding DLM
1961 * lock must exist so that page content must be valid.
1963 * In fast read implementation, the llite speculatively finds and reads pages
1964 * in memory cache. There are three scenarios for fast read:
1965 * - If the page exists and is uptodate, kernel VM will provide the data and
1966 * CLIO won't be intervened;
1967 * - If the page was brought into memory by read ahead, it will be exported
1968 * and read ahead parameters will be updated;
1969 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1970 * it will go back and invoke normal read, i.e., a cl_io will be created
1971 * and DLM lock will be requested.
1973 * POSIX compliance: posix standard states that read is intended to be atomic.
1974 * Lustre read implementation is in line with Linux kernel read implementation
1975 * and neither of them complies with POSIX standard in this matter. Fast read
1976 * doesn't make the situation worse on single node but it may interleave write
1977 * results from multiple nodes due to short read handling in ll_file_aio_read().
1979 * \param env - lu_env
1980 * \param iocb - kiocb from kernel
1981 * \param iter - user space buffers where the data will be copied
1983 * \retval - number of bytes have been read, or error code if error occurred.
1986 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1990 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1993 /* NB: we can't do direct IO for fast read because it will need a lock
1994 * to make IO engine happy. */
1995 if (iocb->ki_filp->f_flags & O_DIRECT)
1998 result = generic_file_read_iter(iocb, iter);
2000 /* If the first page is not in cache, generic_file_aio_read() will be
2001 * returned with -ENODATA. Fall back to full read path.
2002 * See corresponding code in ll_readpage().
2004 * if we raced with page deletion, we might get EIO. Rather than add
2005 * locking to the fast path for this rare case, fall back to the full
2006 * read path. (See vvp_io_read_start() for rest of handling.
2008 if (result == -ENODATA || result == -EIO)
2012 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
2013 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
2014 LPROC_LL_READ_BYTES, result);
2021 * Confine read iter lest read beyond the EOF
2023 * \param iocb [in] kernel iocb
2024 * \param to [in] reader iov_iter
2026 * \retval <0 failure
2028 * \retval >0 @iocb->ki_pos has passed the EOF
2030 static int file_read_confine_iter(struct lu_env *env, struct kiocb *iocb,
2031 struct iov_iter *to)
2033 struct cl_attr *attr = vvp_env_thread_attr(env);
2034 struct file *file = iocb->ki_filp;
2035 struct inode *inode = file_inode(file);
2036 struct ll_inode_info *lli = ll_i2info(inode);
2037 loff_t read_end = iocb->ki_pos + iov_iter_count(to);
2042 cl_object_attr_lock(lli->lli_clob);
2043 rc = cl_object_attr_get(env, lli->lli_clob, attr);
2044 cl_object_attr_unlock(lli->lli_clob);
2048 kms = attr->cat_kms;
2049 /* if read beyond end-of-file, adjust read count */
2050 if (kms > 0 && (iocb->ki_pos >= kms || read_end > kms)) {
2051 rc = ll_glimpse_size(inode);
2055 size = i_size_read(inode);
2056 if (iocb->ki_pos >= size || read_end > size) {
2058 "%s: read [%llu, %llu] over eof, kms %llu, file_size %llu.\n",
2059 file_dentry(file)->d_name.name,
2060 iocb->ki_pos, read_end, kms, size);
2062 if (iocb->ki_pos >= size)
2065 if (read_end > size)
2066 iov_iter_truncate(to, size - iocb->ki_pos);
2074 * Read from a file (through the page cache).
2076 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2079 struct vvp_io_args *args;
2080 struct file *file = iocb->ki_filp;
2084 ktime_t kstart = ktime_get();
2086 bool stale_data = false;
2090 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2091 file_dentry(file)->d_name.name,
2092 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2093 iov_iter_count(to));
2095 if (!iov_iter_count(to))
2098 env = cl_env_get(&refcheck);
2100 RETURN(PTR_ERR(env));
2102 result = file_read_confine_iter(env, iocb, to);
2105 else if (result > 0)
2109 * Currently when PCC read failed, we do not fall back to the
2110 * normal read path, just return the error.
2111 * The resaon is that: for RW-PCC, the file data may be modified
2112 * in the PCC and inconsistent with the data on OSTs (or file
2113 * data has been removed from the Lustre file system), at this
2114 * time, fallback to the normal read path may read the wrong
2116 * TODO: for RO-PCC (readonly PCC), fall back to normal read
2117 * path: read data from data copy on OSTs.
2119 result = pcc_file_read_iter(iocb, to, &cached);
2123 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
2125 result = ll_do_fast_read(iocb, to);
2126 if (result < 0 || iov_iter_count(to) == 0)
2129 args = ll_env_args(env);
2130 args->u.normal.via_iter = to;
2131 args->u.normal.via_iocb = iocb;
2133 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
2134 &iocb->ki_pos, iov_iter_count(to));
2137 else if (result == 0)
2141 cl_env_put(env, &refcheck);
2143 if (stale_data && result > 0) {
2145 * we've reached EOF before the read, the data read are cached
2148 iov_iter_truncate(to, 0);
2153 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2154 file->private_data, iocb->ki_pos, result,
2156 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
2157 ktime_us_delta(ktime_get(), kstart));
2161 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2162 file_dentry(file)->d_name.name,
2163 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2164 iov_iter_count(to));
2170 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
2171 * If a page is already in the page cache and dirty (and some other things -
2172 * See ll_tiny_write_begin for the instantiation of these rules), then we can
2173 * write to it without doing a full I/O, because Lustre already knows about it
2174 * and will write it out. This saves a lot of processing time.
2176 * All writes here are within one page, so exclusion is handled by the page
2177 * lock on the vm page. We do not do tiny writes for writes which touch
2178 * multiple pages because it's very unlikely multiple sequential pages are
2179 * are already dirty.
2181 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
2182 * and are unlikely to be to already dirty pages.
2184 * Attribute updates are important here, we do them in ll_tiny_write_end.
2186 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
2188 ssize_t count = iov_iter_count(iter);
2189 struct file *file = iocb->ki_filp;
2190 struct inode *inode = file_inode(file);
2191 bool lock_inode = !IS_NOSEC(inode);
2196 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2197 * of function for why.
2199 if (count >= PAGE_SIZE ||
2200 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2203 if (unlikely(lock_inode))
2205 result = __generic_file_write_iter(iocb, iter);
2207 if (unlikely(lock_inode))
2208 inode_unlock(inode);
2210 /* If the page is not already dirty, ll_tiny_write_begin returns
2211 * -ENODATA. We continue on to normal write.
2213 if (result == -ENODATA)
2217 ll_heat_add(inode, CIT_WRITE, result);
2218 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2221 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2227 * Write to a file (through the page cache).
2229 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2231 struct vvp_io_args *args;
2233 ssize_t rc_tiny = 0, rc_normal;
2234 struct file *file = iocb->ki_filp;
2237 ktime_t kstart = ktime_get();
2242 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2243 file_dentry(file)->d_name.name,
2244 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2245 iov_iter_count(from));
2247 if (!iov_iter_count(from))
2248 GOTO(out, rc_normal = 0);
2251 * When PCC write failed, we usually do not fall back to the normal
2252 * write path, just return the error. But there is a special case when
2253 * returned error code is -ENOSPC due to running out of space on PCC HSM
2254 * bakcend. At this time, it will fall back to normal I/O path and
2255 * retry the I/O. As the file is in HSM released state, it will restore
2256 * the file data to OSTs first and redo the write again. And the
2257 * restore process will revoke the layout lock and detach the file
2258 * from PCC cache automatically.
2260 result = pcc_file_write_iter(iocb, from, &cached);
2261 if (cached && result != -ENOSPC && result != -EDQUOT)
2262 GOTO(out, rc_normal = result);
2264 /* NB: we can't do direct IO for tiny writes because they use the page
2265 * cache, we can't do sync writes because tiny writes can't flush
2266 * pages, and we can't do append writes because we can't guarantee the
2267 * required DLM locks are held to protect file size.
2269 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2270 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
2271 rc_tiny = ll_do_tiny_write(iocb, from);
2273 /* In case of error, go on and try normal write - Only stop if tiny
2274 * write completed I/O.
2276 if (iov_iter_count(from) == 0)
2277 GOTO(out, rc_normal = rc_tiny);
2279 env = cl_env_get(&refcheck);
2281 RETURN(PTR_ERR(env));
2283 args = ll_env_args(env);
2284 args->u.normal.via_iter = from;
2285 args->u.normal.via_iocb = iocb;
2287 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2288 &iocb->ki_pos, iov_iter_count(from));
2290 /* On success, combine bytes written. */
2291 if (rc_tiny >= 0 && rc_normal > 0)
2292 rc_normal += rc_tiny;
2293 /* On error, only return error from normal write if tiny write did not
2294 * write any bytes. Otherwise return bytes written by tiny write.
2296 else if (rc_tiny > 0)
2297 rc_normal = rc_tiny;
2299 cl_env_put(env, &refcheck);
2301 if (rc_normal > 0) {
2302 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2303 file->private_data, iocb->ki_pos,
2305 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2306 ktime_us_delta(ktime_get(), kstart));
2310 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2311 file_dentry(file)->d_name.name,
2312 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2313 iov_iter_count(from));
2318 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2320 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2322 static int ll_file_get_iov_count(const struct iovec *iov,
2323 unsigned long *nr_segs, size_t *count,
2329 for (seg = 0; seg < *nr_segs; seg++) {
2330 const struct iovec *iv = &iov[seg];
2333 * If any segment has a negative length, or the cumulative
2334 * length ever wraps negative then return -EINVAL.
2337 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2339 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2344 cnt -= iv->iov_len; /* This segment is no good */
2351 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2352 unsigned long nr_segs, loff_t pos)
2359 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2366 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2367 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2368 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2369 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2370 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2372 result = ll_file_read_iter(iocb, &to);
2377 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2380 struct iovec iov = { .iov_base = buf, .iov_len = count };
2389 init_sync_kiocb(&kiocb, file);
2390 kiocb.ki_pos = *ppos;
2391 #ifdef HAVE_KIOCB_KI_LEFT
2392 kiocb.ki_left = count;
2393 #elif defined(HAVE_KI_NBYTES)
2394 kiocb.i_nbytes = count;
2397 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2398 *ppos = kiocb.ki_pos;
2404 * Write to a file (through the page cache).
2407 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2408 unsigned long nr_segs, loff_t pos)
2410 struct iov_iter from;
2415 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2422 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2423 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2424 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2425 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2426 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2428 result = ll_file_write_iter(iocb, &from);
2433 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2434 size_t count, loff_t *ppos)
2436 struct iovec iov = { .iov_base = (void __user *)buf,
2446 init_sync_kiocb(&kiocb, file);
2447 kiocb.ki_pos = *ppos;
2448 #ifdef HAVE_KIOCB_KI_LEFT
2449 kiocb.ki_left = count;
2450 #elif defined(HAVE_KI_NBYTES)
2451 kiocb.ki_nbytes = count;
2454 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2455 *ppos = kiocb.ki_pos;
2459 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2461 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2462 __u64 flags, struct lov_user_md *lum, int lum_size)
2464 struct lookup_intent oit = {
2466 .it_flags = flags | MDS_OPEN_BY_FID,
2471 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2472 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2473 /* this code will only exist for big-endian systems */
2474 lustre_swab_lov_user_md(lum, 0);
2477 ll_inode_size_lock(inode);
2478 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2480 GOTO(out_unlock, rc);
2482 ll_release_openhandle(dentry, &oit);
2485 ll_inode_size_unlock(inode);
2486 ll_intent_release(&oit);
2491 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2492 struct lov_mds_md **lmmp, int *lmm_size,
2493 struct ptlrpc_request **request)
2495 struct ll_sb_info *sbi = ll_i2sbi(inode);
2496 struct mdt_body *body;
2497 struct lov_mds_md *lmm = NULL;
2498 struct ptlrpc_request *req = NULL;
2499 struct md_op_data *op_data;
2504 rc = ll_get_default_mdsize(sbi, &lmmsize);
2508 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2509 strlen(filename), lmmsize,
2510 LUSTRE_OPC_ANY, NULL);
2511 if (IS_ERR(op_data))
2512 RETURN(PTR_ERR(op_data));
2514 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2515 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2516 ll_finish_md_op_data(op_data);
2518 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2523 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2524 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2526 lmmsize = body->mbo_eadatasize;
2528 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2530 GOTO(out, rc = -ENODATA);
2532 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2533 LASSERT(lmm != NULL);
2535 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2536 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2537 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2538 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2539 GOTO(out, rc = -EPROTO);
2542 * This is coming from the MDS, so is probably in
2543 * little endian. We convert it to host endian before
2544 * passing it to userspace.
2546 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2547 int stripe_count = 0;
2549 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2550 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2551 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2552 if (le32_to_cpu(lmm->lmm_pattern) &
2553 LOV_PATTERN_F_RELEASED)
2555 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2557 /* if function called for directory - we should
2558 * avoid swab not existent lsm objects
2560 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2561 S_ISREG(body->mbo_mode))
2562 lustre_swab_lov_user_md_objects(
2563 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2565 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2566 S_ISREG(body->mbo_mode))
2567 lustre_swab_lov_user_md_objects(
2568 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2570 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2571 lustre_swab_lov_comp_md_v1(
2572 (struct lov_comp_md_v1 *)lmm);
2576 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2577 struct lov_comp_md_v1 *comp_v1 = NULL;
2578 struct lov_comp_md_entry_v1 *ent;
2579 struct lov_user_md_v1 *v1;
2583 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2584 /* Dump the striping information */
2585 for (; i < comp_v1->lcm_entry_count; i++) {
2586 ent = &comp_v1->lcm_entries[i];
2587 off = ent->lcme_offset;
2588 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2590 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2591 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2595 * Return valid stripe_count and stripe_size instead of 0 for
2596 * DoM files to avoid divide-by-zero for older userspace that
2597 * calls this ioctl, e.g. lustre ADIO driver.
2599 if (lmm->lmm_stripe_count == 0)
2600 lmm->lmm_stripe_count = 1;
2601 if (lmm->lmm_stripe_size == 0) {
2602 /* Since the first component of the file data is placed
2603 * on the MDT for faster access, the stripe_size of the
2604 * second one is always that applications which are
2607 if (lmm->lmm_pattern == LOV_PATTERN_MDT)
2608 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2610 i = comp_v1->lcm_entry_count > 1 ?
2611 comp_v1->lcm_entry_count - 1 : 0;
2612 ent = &comp_v1->lcm_entries[i];
2613 off = ent->lcme_offset;
2614 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2615 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2620 *lmm_size = lmmsize;
2625 static int ll_lov_setea(struct inode *inode, struct file *file,
2628 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2629 struct lov_user_md *lump;
2630 int lum_size = sizeof(*lump) + sizeof(struct lov_user_ost_data);
2634 if (!capable(CAP_SYS_ADMIN))
2637 OBD_ALLOC_LARGE(lump, lum_size);
2641 if (copy_from_user(lump, arg, lum_size))
2642 GOTO(out_lump, rc = -EFAULT);
2644 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2646 cl_lov_delay_create_clear(&file->f_flags);
2649 OBD_FREE_LARGE(lump, lum_size);
2653 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2660 /* exit before doing any work if pointer is bad */
2661 if (unlikely(!ll_access_ok(lum, sizeof(struct lov_user_md))))
2664 env = cl_env_get(&refcheck);
2666 RETURN(PTR_ERR(env));
2668 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2669 cl_env_put(env, &refcheck);
2673 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2676 struct lov_user_md __user *lum = arg;
2677 struct lov_user_md *klum;
2679 __u64 flags = FMODE_WRITE;
2682 rc = ll_copy_user_md(lum, &klum);
2687 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2692 rc = put_user(0, &lum->lmm_stripe_count);
2696 rc = ll_layout_refresh(inode, &gen);
2700 rc = ll_file_getstripe(inode, arg, lum_size);
2701 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2702 ll_i2info(inode)->lli_clob) {
2703 struct iattr attr = { 0 };
2705 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2706 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2709 cl_lov_delay_create_clear(&file->f_flags);
2712 OBD_FREE_LARGE(klum, lum_size);
2718 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2720 struct ll_inode_info *lli = ll_i2info(inode);
2721 struct cl_object *obj = lli->lli_clob;
2722 struct ll_file_data *fd = file->private_data;
2723 struct ll_grouplock grouplock;
2728 CWARN("group id for group lock must not be 0\n");
2732 if (ll_file_nolock(file))
2733 RETURN(-EOPNOTSUPP);
2735 if (file->f_flags & O_NONBLOCK) {
2736 if (!mutex_trylock(&lli->lli_group_mutex))
2739 mutex_lock(&lli->lli_group_mutex);
2742 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2743 CWARN("group lock already existed with gid %lu\n",
2744 fd->fd_grouplock.lg_gid);
2745 GOTO(out, rc = -EINVAL);
2747 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2748 if (file->f_flags & O_NONBLOCK)
2749 GOTO(out, rc = -EAGAIN);
2750 mutex_unlock(&lli->lli_group_mutex);
2751 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2752 GOTO(retry, rc = 0);
2754 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2757 * XXX: group lock needs to protect all OST objects while PFL
2758 * can add new OST objects during the IO, so we'd instantiate
2759 * all OST objects before getting its group lock.
2764 struct cl_layout cl = {
2765 .cl_is_composite = false,
2767 struct lu_extent ext = {
2769 .e_end = OBD_OBJECT_EOF,
2772 env = cl_env_get(&refcheck);
2774 GOTO(out, rc = PTR_ERR(env));
2776 rc = cl_object_layout_get(env, obj, &cl);
2777 if (rc >= 0 && cl.cl_is_composite)
2778 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2781 cl_env_put(env, &refcheck);
2786 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2787 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2792 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2793 fd->fd_grouplock = grouplock;
2794 if (lli->lli_group_users == 0)
2795 lli->lli_group_gid = grouplock.lg_gid;
2796 lli->lli_group_users++;
2798 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2800 mutex_unlock(&lli->lli_group_mutex);
2805 static int ll_put_grouplock(struct inode *inode, struct file *file,
2808 struct ll_inode_info *lli = ll_i2info(inode);
2809 struct ll_file_data *fd = file->private_data;
2810 struct ll_grouplock grouplock;
2814 mutex_lock(&lli->lli_group_mutex);
2815 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2816 CWARN("no group lock held\n");
2817 GOTO(out, rc = -EINVAL);
2820 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2822 if (fd->fd_grouplock.lg_gid != arg) {
2823 CWARN("group lock %lu doesn't match current id %lu\n",
2824 arg, fd->fd_grouplock.lg_gid);
2825 GOTO(out, rc = -EINVAL);
2828 grouplock = fd->fd_grouplock;
2829 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2830 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2832 cl_put_grouplock(&grouplock);
2834 lli->lli_group_users--;
2835 if (lli->lli_group_users == 0) {
2836 lli->lli_group_gid = 0;
2837 wake_up_var(&lli->lli_group_users);
2839 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2842 mutex_unlock(&lli->lli_group_mutex);
2848 * Close inode open handle
2850 * \param dentry [in] dentry which contains the inode
2851 * \param it [in,out] intent which contains open info and result
2854 * \retval <0 failure
2856 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2858 struct inode *inode = dentry->d_inode;
2859 struct obd_client_handle *och;
2865 /* Root ? Do nothing. */
2866 if (is_root_inode(inode))
2869 /* No open handle to close? Move away */
2870 if (!it_disposition(it, DISP_OPEN_OPEN))
2873 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2875 OBD_ALLOC(och, sizeof(*och));
2877 GOTO(out, rc = -ENOMEM);
2879 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2883 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2885 /* this one is in place of ll_file_open */
2886 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2887 ptlrpc_req_finished(it->it_request);
2888 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2894 * Get size for inode for which FIEMAP mapping is requested.
2895 * Make the FIEMAP get_info call and returns the result.
2896 * \param fiemap kernel buffer to hold extens
2897 * \param num_bytes kernel buffer size
2899 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2905 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2908 /* Checks for fiemap flags */
2909 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2910 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2914 /* Check for FIEMAP_FLAG_SYNC */
2915 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2916 rc = filemap_fdatawrite(inode->i_mapping);
2921 env = cl_env_get(&refcheck);
2923 RETURN(PTR_ERR(env));
2925 if (i_size_read(inode) == 0) {
2926 rc = ll_glimpse_size(inode);
2931 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLPROJID;
2932 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2933 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2935 /* If filesize is 0, then there would be no objects for mapping */
2936 if (fmkey.lfik_oa.o_size == 0) {
2937 fiemap->fm_mapped_extents = 0;
2941 fmkey.lfik_fiemap = *fiemap;
2943 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2944 &fmkey, fiemap, &num_bytes);
2946 cl_env_put(env, &refcheck);
2950 static int fid2path_for_enc_file(struct inode *parent, char *gfpath,
2953 struct dentry *de = NULL, *de_parent = d_find_any_alias(parent);
2954 struct llcrypt_str lltr = LLTR_INIT(NULL, 0);
2955 struct llcrypt_str de_name;
2956 char *p, *ptr = gfpath;
2957 size_t len = 0, len_orig = 0;
2958 int enckey = -1, nameenc = -1;
2962 while ((p = strsep(&gfpath, "/")) != NULL) {
2970 len_orig = strlen(p);
2972 rc = sscanf(p, "["SFID"]", RFID(&fid));
2974 p = strchr(p, ']') + 1;
2980 if (!IS_ENCRYPTED(parent)) {
2981 if (gfpathlen < len + 1) {
2986 memmove(ptr, p, len);
2990 gfpathlen -= len + 1;
2994 /* From here, we know parent is encrypted */
2997 rc = llcrypt_prepare_readdir(parent);
2998 if (rc && rc != -ENOKEY) {
3005 if (llcrypt_has_encryption_key(parent))
3011 llcrypt_policy_has_filename_enc(parent);
3014 /* Even if names are not encrypted, we still need to call
3015 * ll_fname_disk_to_usr in order to decode names as they are
3016 * coming from the wire.
3018 rc = llcrypt_fname_alloc_buffer(parent, NAME_MAX + 1, &lltr);
3026 rc = ll_fname_disk_to_usr(parent, 0, 0, &de_name,
3029 llcrypt_fname_free_buffer(&lltr);
3033 lltr.name[lltr.len] = '\0';
3035 if (lltr.len <= len_orig && gfpathlen >= lltr.len + 1) {
3036 memcpy(ptr, lltr.name, lltr.len);
3041 gfpathlen -= lltr.len + 1;
3045 llcrypt_fname_free_buffer(&lltr);
3047 if (rc == -EOVERFLOW) {
3054 /* We reached the end of the string, which means
3055 * we are dealing with the last component in the path.
3056 * So save a useless lookup and exit.
3062 if (enckey == 0 || nameenc == 0)
3066 de = lookup_one_len(p, de_parent, len);
3067 inode_unlock(parent);
3068 if (IS_ERR_OR_NULL(de) || !de->d_inode) {
3074 parent = de->d_inode;
3081 if (!IS_ERR_OR_NULL(de))
3086 int __ll_fid2path(struct inode *inode, struct getinfo_fid2path *gfout,
3087 size_t outsize, __u32 pathlen_orig)
3089 struct obd_export *exp = ll_i2mdexp(inode);
3092 /* Append root FID after gfout to let MDT know the root FID so that
3093 * it can lookup the correct path, this is mainly for fileset.
3094 * old server without fileset mount support will ignore this.
3096 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
3098 /* Call mdc_iocontrol */
3099 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
3101 if (!rc && gfout->gf_pathlen && gfout->gf_u.gf_path[0] == '/') {
3102 /* by convention, server side (mdt_path_current()) puts
3103 * a leading '/' to tell client that we are dealing with
3106 rc = fid2path_for_enc_file(inode, gfout->gf_u.gf_path,
3108 if (!rc && strlen(gfout->gf_u.gf_path) > pathlen_orig)
3115 int ll_fid2path(struct inode *inode, void __user *arg)
3117 const struct getinfo_fid2path __user *gfin = arg;
3118 __u32 pathlen, pathlen_orig;
3119 struct getinfo_fid2path *gfout;
3125 if (!capable(CAP_DAC_READ_SEARCH) &&
3126 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3129 /* Only need to get the buflen */
3130 if (get_user(pathlen, &gfin->gf_pathlen))
3133 if (pathlen > PATH_MAX)
3135 pathlen_orig = pathlen;
3138 outsize = sizeof(*gfout) + pathlen;
3139 OBD_ALLOC(gfout, outsize);
3143 if (copy_from_user(gfout, arg, sizeof(*gfout)))
3144 GOTO(gf_free, rc = -EFAULT);
3146 gfout->gf_pathlen = pathlen;
3147 rc = __ll_fid2path(inode, gfout, outsize, pathlen_orig);
3151 if (copy_to_user(arg, gfout, sizeof(*gfout) + pathlen_orig))
3155 OBD_FREE(gfout, outsize);
3156 if (rc == -ENAMETOOLONG) {
3157 pathlen += PATH_MAX;
3164 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
3166 struct cl_object *obj = ll_i2info(inode)->lli_clob;
3174 ioc->idv_version = 0;
3175 ioc->idv_layout_version = UINT_MAX;
3177 /* If no file object initialized, we consider its version is 0. */
3181 env = cl_env_get(&refcheck);
3183 RETURN(PTR_ERR(env));
3185 io = vvp_env_thread_io(env);
3187 io->u.ci_data_version.dv_data_version = 0;
3188 io->u.ci_data_version.dv_layout_version = UINT_MAX;
3189 io->u.ci_data_version.dv_flags = ioc->idv_flags;
3192 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
3193 result = cl_io_loop(env, io);
3195 result = io->ci_result;
3197 ioc->idv_version = io->u.ci_data_version.dv_data_version;
3198 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
3200 cl_io_fini(env, io);
3202 if (unlikely(io->ci_need_restart))
3205 cl_env_put(env, &refcheck);
3211 * Read the data_version for inode.
3213 * This value is computed using stripe object version on OST.
3214 * Version is computed using server side locking.
3216 * @param flags if do sync on the OST side;
3218 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
3219 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
3221 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
3223 struct ioc_data_version ioc = { .idv_flags = flags };
3226 rc = ll_ioc_data_version(inode, &ioc);
3228 *data_version = ioc.idv_version;
3234 * Trigger a HSM release request for the provided inode.
3236 int ll_hsm_release(struct inode *inode)
3239 struct obd_client_handle *och = NULL;
3240 __u64 data_version = 0;
3246 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
3247 ll_i2sbi(inode)->ll_fsname,
3248 PFID(&ll_i2info(inode)->lli_fid));
3250 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
3252 GOTO(out, rc = PTR_ERR(och));
3254 /* Grab latest data_version and [am]time values */
3255 rc = ll_data_version(inode, &data_version,
3256 LL_DV_WR_FLUSH | LL_DV_SZ_UPDATE);
3260 env = cl_env_get(&refcheck);
3262 GOTO(out, rc = PTR_ERR(env));
3264 rc = ll_merge_attr(env, inode);
3265 cl_env_put(env, &refcheck);
3267 /* If error happen, we have the wrong size for a file.
3273 /* Release the file.
3274 * NB: lease lock handle is released in mdc_hsm_release_pack() because
3275 * we still need it to pack l_remote_handle to MDT. */
3276 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
3282 if (och != NULL && !IS_ERR(och)) /* close the file */
3283 ll_lease_close(och, inode, NULL);
3288 struct ll_swap_stack {
3291 struct inode *inode1;
3292 struct inode *inode2;
3297 static int ll_swap_layouts(struct file *file1, struct file *file2,
3298 struct lustre_swap_layouts *lsl)
3300 struct mdc_swap_layouts msl;
3301 struct md_op_data *op_data;
3304 struct ll_swap_stack *llss = NULL;
3307 OBD_ALLOC_PTR(llss);
3311 llss->inode1 = file_inode(file1);
3312 llss->inode2 = file_inode(file2);
3314 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
3318 /* we use 2 bool because it is easier to swap than 2 bits */
3319 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
3320 llss->check_dv1 = true;
3322 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
3323 llss->check_dv2 = true;
3325 /* we cannot use lsl->sl_dvX directly because we may swap them */
3326 llss->dv1 = lsl->sl_dv1;
3327 llss->dv2 = lsl->sl_dv2;
3329 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
3330 if (rc == 0) /* same file, done! */
3333 if (rc < 0) { /* sequentialize it */
3334 swap(llss->inode1, llss->inode2);
3336 swap(llss->dv1, llss->dv2);
3337 swap(llss->check_dv1, llss->check_dv2);
3341 if (gid != 0) { /* application asks to flush dirty cache */
3342 rc = ll_get_grouplock(llss->inode1, file1, gid);
3346 rc = ll_get_grouplock(llss->inode2, file2, gid);
3348 ll_put_grouplock(llss->inode1, file1, gid);
3353 /* ultimate check, before swaping the layouts we check if
3354 * dataversion has changed (if requested) */
3355 if (llss->check_dv1) {
3356 rc = ll_data_version(llss->inode1, &dv, 0);
3359 if (dv != llss->dv1)
3360 GOTO(putgl, rc = -EAGAIN);
3363 if (llss->check_dv2) {
3364 rc = ll_data_version(llss->inode2, &dv, 0);
3367 if (dv != llss->dv2)
3368 GOTO(putgl, rc = -EAGAIN);
3371 /* struct md_op_data is used to send the swap args to the mdt
3372 * only flags is missing, so we use struct mdc_swap_layouts
3373 * through the md_op_data->op_data */
3374 /* flags from user space have to be converted before they are send to
3375 * server, no flag is sent today, they are only used on the client */
3378 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3379 0, LUSTRE_OPC_ANY, &msl);
3380 if (IS_ERR(op_data))
3381 GOTO(free, rc = PTR_ERR(op_data));
3383 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3384 sizeof(*op_data), op_data, NULL);
3385 ll_finish_md_op_data(op_data);
3392 ll_put_grouplock(llss->inode2, file2, gid);
3393 ll_put_grouplock(llss->inode1, file1, gid);
3403 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3405 struct obd_export *exp = ll_i2mdexp(inode);
3406 struct md_op_data *op_data;
3410 /* Detect out-of range masks */
3411 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3414 /* Non-root users are forbidden to set or clear flags which are
3415 * NOT defined in HSM_USER_MASK. */
3416 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3417 !capable(CAP_SYS_ADMIN))
3420 if (!exp_connect_archive_id_array(exp)) {
3421 /* Detect out-of range archive id */
3422 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3423 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3427 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3428 LUSTRE_OPC_ANY, hss);
3429 if (IS_ERR(op_data))
3430 RETURN(PTR_ERR(op_data));
3432 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3435 ll_finish_md_op_data(op_data);
3440 static int ll_hsm_import(struct inode *inode, struct file *file,
3441 struct hsm_user_import *hui)
3443 struct hsm_state_set *hss = NULL;
3444 struct iattr *attr = NULL;
3448 if (!S_ISREG(inode->i_mode))
3454 GOTO(out, rc = -ENOMEM);
3456 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3457 hss->hss_archive_id = hui->hui_archive_id;
3458 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3459 rc = ll_hsm_state_set(inode, hss);
3463 OBD_ALLOC_PTR(attr);
3465 GOTO(out, rc = -ENOMEM);
3467 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3468 attr->ia_mode |= S_IFREG;
3469 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3470 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3471 attr->ia_size = hui->hui_size;
3472 attr->ia_mtime.tv_sec = hui->hui_mtime;
3473 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3474 attr->ia_atime.tv_sec = hui->hui_atime;
3475 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3477 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3478 ATTR_UID | ATTR_GID |
3479 ATTR_MTIME | ATTR_MTIME_SET |
3480 ATTR_ATIME | ATTR_ATIME_SET;
3484 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3488 inode_unlock(inode);
3500 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3502 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3503 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3506 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3508 struct inode *inode = file_inode(file);
3510 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3511 ATTR_MTIME | ATTR_MTIME_SET |
3514 .tv_sec = lfu->lfu_atime_sec,
3515 .tv_nsec = lfu->lfu_atime_nsec,
3518 .tv_sec = lfu->lfu_mtime_sec,
3519 .tv_nsec = lfu->lfu_mtime_nsec,
3522 .tv_sec = lfu->lfu_ctime_sec,
3523 .tv_nsec = lfu->lfu_ctime_nsec,
3529 if (!capable(CAP_SYS_ADMIN))
3532 if (!S_ISREG(inode->i_mode))
3536 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3538 inode_unlock(inode);
3543 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3546 case MODE_READ_USER:
3548 case MODE_WRITE_USER:
3555 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3557 /* Used to allow the upper layers of the client to request an LDLM lock
3558 * without doing an actual read or write.
3560 * Used for ladvise lockahead to manually request specific locks.
3562 * \param[in] file file this ladvise lock request is on
3563 * \param[in] ladvise ladvise struct describing this lock request
3565 * \retval 0 success, no detailed result available (sync requests
3566 * and requests sent to the server [not handled locally]
3567 * cannot return detailed results)
3568 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3569 * see definitions for details.
3570 * \retval negative negative errno on error
3572 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3574 struct lu_env *env = NULL;
3575 struct cl_io *io = NULL;
3576 struct cl_lock *lock = NULL;
3577 struct cl_lock_descr *descr = NULL;
3578 struct dentry *dentry = file->f_path.dentry;
3579 struct inode *inode = dentry->d_inode;
3580 enum cl_lock_mode cl_mode;
3581 off_t start = ladvise->lla_start;
3582 off_t end = ladvise->lla_end;
3589 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3590 dentry, dentry->d_inode,
3591 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3594 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3596 GOTO(out, result = cl_mode);
3598 /* Get IO environment */
3599 result = cl_io_get(inode, &env, &io, &refcheck);
3603 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3606 * nothing to do for this io. This currently happens when
3607 * stripe sub-object's are not yet created.
3609 result = io->ci_result;
3610 } else if (result == 0) {
3611 lock = vvp_env_lock(env);
3612 descr = &lock->cll_descr;
3614 descr->cld_obj = io->ci_obj;
3615 /* Convert byte offsets to pages */
3616 descr->cld_start = start >> PAGE_SHIFT;
3617 descr->cld_end = end >> PAGE_SHIFT;
3618 descr->cld_mode = cl_mode;
3619 /* CEF_MUST is used because we do not want to convert a
3620 * lockahead request to a lockless lock */
3621 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3623 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3624 descr->cld_enq_flags |= CEF_SPECULATIVE;
3626 result = cl_lock_request(env, io, lock);
3628 /* On success, we need to release the lock */
3630 cl_lock_release(env, lock);
3632 cl_io_fini(env, io);
3633 cl_env_put(env, &refcheck);
3635 /* -ECANCELED indicates a matching lock with a different extent
3636 * was already present, and -EEXIST indicates a matching lock
3637 * on exactly the same extent was already present.
3638 * We convert them to positive values for userspace to make
3639 * recognizing true errors easier.
3640 * Note we can only return these detailed results on async requests,
3641 * as sync requests look the same as i/o requests for locking. */
3642 if (result == -ECANCELED)
3643 result = LLA_RESULT_DIFFERENT;
3644 else if (result == -EEXIST)
3645 result = LLA_RESULT_SAME;
3650 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3652 static int ll_ladvise_sanity(struct inode *inode,
3653 struct llapi_lu_ladvise *ladvise)
3655 struct ll_sb_info *sbi = ll_i2sbi(inode);
3656 enum lu_ladvise_type advice = ladvise->lla_advice;
3657 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3658 * be in the first 32 bits of enum ladvise_flags */
3659 __u32 flags = ladvise->lla_peradvice_flags;
3660 /* 3 lines at 80 characters per line, should be plenty */
3663 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3666 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3667 sbi->ll_fsname, advice,
3668 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3672 /* Per-advice checks */
3674 case LU_LADVISE_LOCKNOEXPAND:
3675 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3677 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3678 "rc = %d\n", sbi->ll_fsname, flags,
3679 ladvise_names[advice], rc);
3683 case LU_LADVISE_LOCKAHEAD:
3684 /* Currently only READ and WRITE modes can be requested */
3685 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3686 ladvise->lla_lockahead_mode == 0) {
3688 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3689 "rc = %d\n", sbi->ll_fsname,
3690 ladvise->lla_lockahead_mode,
3691 ladvise_names[advice], rc);
3695 case LU_LADVISE_WILLREAD:
3696 case LU_LADVISE_DONTNEED:
3698 /* Note fall through above - These checks apply to all advices
3699 * except LOCKNOEXPAND */
3700 if (flags & ~LF_DEFAULT_MASK) {
3702 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3703 "rc = %d\n", sbi->ll_fsname, flags,
3704 ladvise_names[advice], rc);
3707 if (ladvise->lla_start >= ladvise->lla_end) {
3709 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3710 "for %s: rc = %d\n", sbi->ll_fsname,
3711 ladvise->lla_start, ladvise->lla_end,
3712 ladvise_names[advice], rc);
3724 * Give file access advices
3726 * The ladvise interface is similar to Linux fadvise() system call, except it
3727 * forwards the advices directly from Lustre client to server. The server side
3728 * codes will apply appropriate read-ahead and caching techniques for the
3729 * corresponding files.
3731 * A typical workload for ladvise is e.g. a bunch of different clients are
3732 * doing small random reads of a file, so prefetching pages into OSS cache
3733 * with big linear reads before the random IO is a net benefit. Fetching
3734 * all that data into each client cache with fadvise() may not be, due to
3735 * much more data being sent to the client.
3737 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3738 struct llapi_lu_ladvise *ladvise)
3742 struct cl_ladvise_io *lio;
3747 env = cl_env_get(&refcheck);
3749 RETURN(PTR_ERR(env));
3751 io = vvp_env_thread_io(env);
3752 io->ci_obj = ll_i2info(inode)->lli_clob;
3754 /* initialize parameters for ladvise */
3755 lio = &io->u.ci_ladvise;
3756 lio->li_start = ladvise->lla_start;
3757 lio->li_end = ladvise->lla_end;
3758 lio->li_fid = ll_inode2fid(inode);
3759 lio->li_advice = ladvise->lla_advice;
3760 lio->li_flags = flags;
3762 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3763 rc = cl_io_loop(env, io);
3767 cl_io_fini(env, io);
3768 cl_env_put(env, &refcheck);
3772 static int ll_lock_noexpand(struct file *file, int flags)
3774 struct ll_file_data *fd = file->private_data;
3776 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3781 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3784 struct fsxattr fsxattr;
3786 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3789 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3790 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3791 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3792 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3793 if (copy_to_user(uarg, &fsxattr, sizeof(fsxattr)))
3799 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3803 * Project Quota ID state is only allowed to change from within the init
3804 * namespace. Enforce that restriction only if we are trying to change
3805 * the quota ID state. Everything else is allowed in user namespaces.
3807 if (current_user_ns() == &init_user_ns) {
3809 * Caller is allowed to change the project ID. if it is being
3810 * changed, make sure that the new value is valid.
3812 if (ll_i2info(inode)->lli_projid != projid &&
3813 !projid_valid(make_kprojid(&init_user_ns, projid)))
3819 if (ll_i2info(inode)->lli_projid != projid)
3822 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3823 if (!(xflags & FS_XFLAG_PROJINHERIT))
3826 if (xflags & FS_XFLAG_PROJINHERIT)
3833 static int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3835 struct ptlrpc_request *req = NULL;
3836 struct md_op_data *op_data;
3837 struct cl_object *obj;
3838 unsigned int inode_flags;
3841 CDEBUG(D_QUOTA, DFID" xflags=%x projid=%u\n",
3842 PFID(ll_inode2fid(inode)), xflags, projid);
3843 rc = ll_ioctl_check_project(inode, xflags, projid);
3847 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3848 LUSTRE_OPC_ANY, NULL);
3849 if (IS_ERR(op_data))
3850 RETURN(PTR_ERR(op_data));
3852 inode_flags = ll_xflags_to_inode_flags(xflags);
3853 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3854 if (xflags & FS_XFLAG_PROJINHERIT)
3855 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3857 /* pass projid to md_op_data */
3858 op_data->op_projid = projid;
3860 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3861 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3862 ptlrpc_req_finished(req);
3864 GOTO(out_fsxattr, rc);
3865 ll_update_inode_flags(inode, op_data->op_attr_flags);
3867 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3868 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3869 GOTO(out_fsxattr, rc);
3871 obj = ll_i2info(inode)->lli_clob;
3873 struct iattr attr = { 0 };
3875 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3879 ll_finish_md_op_data(op_data);
3883 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3886 struct fsxattr fsxattr;
3890 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3893 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3894 fsxattr.fsx_projid));
3897 int ll_ioctl_project(struct file *file, unsigned int cmd, void __user *uarg)
3899 struct lu_project lu_project;
3900 struct dentry *dentry = file_dentry(file);
3901 struct inode *inode = file_inode(file);
3902 struct dentry *child_dentry = NULL;
3903 int rc = 0, name_len;
3905 if (copy_from_user(&lu_project, uarg, sizeof(lu_project)))
3908 /* apply child dentry if name is valid */
3909 name_len = strnlen(lu_project.project_name, NAME_MAX);
3910 if (name_len > 0 && name_len <= NAME_MAX) {
3912 child_dentry = lookup_one_len(lu_project.project_name,
3914 inode_unlock(inode);
3915 if (IS_ERR(child_dentry)) {
3916 rc = PTR_ERR(child_dentry);
3919 inode = child_dentry->d_inode;
3924 } else if (name_len > NAME_MAX) {
3929 switch (lu_project.project_type) {
3930 case LU_PROJECT_SET:
3931 rc = ll_set_project(inode, lu_project.project_xflags,
3932 lu_project.project_id);
3934 case LU_PROJECT_GET:
3935 lu_project.project_xflags =
3936 ll_inode_flags_to_xflags(inode->i_flags);
3937 if (test_bit(LLIF_PROJECT_INHERIT,
3938 &ll_i2info(inode)->lli_flags))
3939 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3940 lu_project.project_id = ll_i2info(inode)->lli_projid;
3941 if (copy_to_user(uarg, &lu_project, sizeof(lu_project))) {
3951 if (!IS_ERR_OR_NULL(child_dentry))
3956 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3959 struct inode *inode = file_inode(file);
3960 struct ll_file_data *fd = file->private_data;
3961 struct ll_inode_info *lli = ll_i2info(inode);
3962 struct obd_client_handle *och = NULL;
3963 struct split_param sp;
3964 struct pcc_param param;
3965 bool lease_broken = false;
3967 enum mds_op_bias bias = 0;
3969 struct file *layout_file = NULL;
3971 size_t data_size = 0;
3972 bool attached = false;
3977 mutex_lock(&lli->lli_och_mutex);
3978 if (fd->fd_lease_och != NULL) {
3979 och = fd->fd_lease_och;
3980 fd->fd_lease_och = NULL;
3982 mutex_unlock(&lli->lli_och_mutex);
3987 fmode = och->och_flags;
3989 switch (ioc->lil_flags) {
3990 case LL_LEASE_RESYNC_DONE:
3991 if (ioc->lil_count > IOC_IDS_MAX)
3992 GOTO(out_lease_close, rc = -EINVAL);
3994 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3995 OBD_ALLOC(data, data_size);
3997 GOTO(out_lease_close, rc = -ENOMEM);
3999 if (copy_from_user(data, uarg, data_size))
4000 GOTO(out_lease_close, rc = -EFAULT);
4002 bias = MDS_CLOSE_RESYNC_DONE;
4004 case LL_LEASE_LAYOUT_MERGE:
4005 if (ioc->lil_count != 1)
4006 GOTO(out_lease_close, rc = -EINVAL);
4008 uarg += sizeof(*ioc);
4009 if (copy_from_user(&fdv, uarg, sizeof(fdv)))
4010 GOTO(out_lease_close, rc = -EFAULT);
4012 layout_file = fget(fdv);
4014 GOTO(out_lease_close, rc = -EBADF);
4016 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
4017 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
4018 GOTO(out_lease_close, rc = -EPERM);
4020 data = file_inode(layout_file);
4021 bias = MDS_CLOSE_LAYOUT_MERGE;
4023 case LL_LEASE_LAYOUT_SPLIT: {
4026 if (ioc->lil_count != 2)
4027 GOTO(out_lease_close, rc = -EINVAL);
4029 uarg += sizeof(*ioc);
4030 if (copy_from_user(&fdv, uarg, sizeof(fdv)))
4031 GOTO(out_lease_close, rc = -EFAULT);
4033 uarg += sizeof(fdv);
4034 if (copy_from_user(&mirror_id, uarg, sizeof(mirror_id)))
4035 GOTO(out_lease_close, rc = -EFAULT);
4036 if (mirror_id >= MIRROR_ID_NEG)
4037 GOTO(out_lease_close, rc = -EINVAL);
4039 layout_file = fget(fdv);
4041 GOTO(out_lease_close, rc = -EBADF);
4043 /* if layout_file == file, it means to destroy the mirror */
4044 sp.sp_inode = file_inode(layout_file);
4045 sp.sp_mirror_id = (__u16)mirror_id;
4047 bias = MDS_CLOSE_LAYOUT_SPLIT;
4050 case LL_LEASE_PCC_ATTACH:
4051 if (ioc->lil_count != 1)
4054 if (IS_ENCRYPTED(inode))
4055 RETURN(-EOPNOTSUPP);
4057 uarg += sizeof(*ioc);
4058 if (copy_from_user(¶m.pa_archive_id, uarg, sizeof(__u32)))
4059 GOTO(out_lease_close, rc2 = -EFAULT);
4061 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
4063 GOTO(out_lease_close, rc2);
4066 /* Grab latest data version */
4067 rc2 = ll_data_version(inode, ¶m.pa_data_version,
4070 GOTO(out_lease_close, rc2);
4073 bias = MDS_PCC_ATTACH;
4076 /* without close intent */
4081 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
4085 rc = ll_lease_och_release(inode, file);
4094 if (ioc->lil_flags == LL_LEASE_RESYNC_DONE && data)
4095 OBD_FREE(data, data_size);
4100 if (ioc->lil_flags == LL_LEASE_PCC_ATTACH) {
4103 rc = pcc_readwrite_attach_fini(file, inode,
4104 param.pa_layout_gen,
4109 ll_layout_refresh(inode, &fd->fd_layout_version);
4112 rc = ll_lease_type_from_fmode(fmode);
4116 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
4119 struct inode *inode = file_inode(file);
4120 struct ll_inode_info *lli = ll_i2info(inode);
4121 struct ll_file_data *fd = file->private_data;
4122 struct obd_client_handle *och = NULL;
4123 __u64 open_flags = 0;
4129 switch (ioc->lil_mode) {
4130 case LL_LEASE_WRLCK:
4131 if (!(file->f_mode & FMODE_WRITE))
4133 fmode = FMODE_WRITE;
4135 case LL_LEASE_RDLCK:
4136 if (!(file->f_mode & FMODE_READ))
4140 case LL_LEASE_UNLCK:
4141 RETURN(ll_file_unlock_lease(file, ioc, uarg));
4146 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
4148 /* apply for lease */
4149 if (ioc->lil_flags & LL_LEASE_RESYNC)
4150 open_flags = MDS_OPEN_RESYNC;
4151 och = ll_lease_open(inode, file, fmode, open_flags);
4153 RETURN(PTR_ERR(och));
4155 if (ioc->lil_flags & LL_LEASE_RESYNC) {
4156 rc = ll_lease_file_resync(och, inode, uarg);
4158 ll_lease_close(och, inode, NULL);
4161 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
4163 ll_lease_close(och, inode, NULL);
4169 mutex_lock(&lli->lli_och_mutex);
4170 if (fd->fd_lease_och == NULL) {
4171 fd->fd_lease_och = och;
4174 mutex_unlock(&lli->lli_och_mutex);
4176 /* impossible now that only excl is supported for now */
4177 ll_lease_close(och, inode, &lease_broken);
4183 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
4185 struct ll_inode_info *lli = ll_i2info(inode);
4186 struct ll_sb_info *sbi = ll_i2sbi(inode);
4187 __u64 now = ktime_get_real_seconds();
4190 spin_lock(&lli->lli_heat_lock);
4191 heat->lh_flags = lli->lli_heat_flags;
4192 for (i = 0; i < heat->lh_count; i++)
4193 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
4194 now, sbi->ll_heat_decay_weight,
4195 sbi->ll_heat_period_second);
4196 spin_unlock(&lli->lli_heat_lock);
4199 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
4201 struct ll_inode_info *lli = ll_i2info(inode);
4204 spin_lock(&lli->lli_heat_lock);
4205 if (flags & LU_HEAT_FLAG_CLEAR)
4206 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
4208 if (flags & LU_HEAT_FLAG_OFF)
4209 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
4211 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
4213 spin_unlock(&lli->lli_heat_lock);
4219 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4221 struct inode *inode = file_inode(file);
4222 struct ll_file_data *fd = file->private_data;
4223 void __user *uarg = (void __user *)arg;
4227 CDEBUG(D_VFSTRACE|D_IOCTL, "VFS Op:inode="DFID"(%pK) cmd=%x arg=%lx\n",
4228 PFID(ll_inode2fid(inode)), inode, cmd, arg);
4229 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
4231 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
4232 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
4235 /* can't do a generic karg == NULL check here, since it is too noisy and
4236 * we need to return -ENOTTY for unsupported ioctls instead of -EINVAL.
4239 case LL_IOC_GETFLAGS:
4240 /* Get the current value of the file flags */
4241 return put_user(fd->fd_flags, (int __user *)arg);
4242 case LL_IOC_SETFLAGS:
4243 case LL_IOC_CLRFLAGS:
4244 /* Set or clear specific file flags */
4245 /* XXX This probably needs checks to ensure the flags are
4246 * not abused, and to handle any flag side effects.
4248 if (get_user(flags, (int __user *)arg))
4251 if (cmd == LL_IOC_SETFLAGS) {
4252 if ((flags & LL_FILE_IGNORE_LOCK) &&
4253 !(file->f_flags & O_DIRECT)) {
4255 CERROR("%s: unable to disable locking on non-O_DIRECT file "DFID": rc = %d\n",
4256 current->comm, PFID(ll_inode2fid(inode)),
4261 fd->fd_flags |= flags;
4263 fd->fd_flags &= ~flags;
4266 case LL_IOC_LOV_SETSTRIPE:
4267 case LL_IOC_LOV_SETSTRIPE_NEW:
4268 RETURN(ll_lov_setstripe(inode, file, uarg));
4269 case LL_IOC_LOV_SETEA:
4270 RETURN(ll_lov_setea(inode, file, uarg));
4271 case LL_IOC_LOV_SWAP_LAYOUTS: {
4273 struct lustre_swap_layouts lsl;
4275 if (copy_from_user(&lsl, uarg, sizeof(lsl)))
4278 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
4281 file2 = fget(lsl.sl_fd);
4285 /* O_WRONLY or O_RDWR */
4286 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
4287 GOTO(out, rc = -EPERM);
4289 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
4290 struct obd_client_handle *och = NULL;
4291 struct ll_inode_info *lli;
4292 struct inode *inode2;
4294 lli = ll_i2info(inode);
4295 mutex_lock(&lli->lli_och_mutex);
4296 if (fd->fd_lease_och != NULL) {
4297 och = fd->fd_lease_och;
4298 fd->fd_lease_och = NULL;
4300 mutex_unlock(&lli->lli_och_mutex);
4302 GOTO(out, rc = -ENOLCK);
4303 inode2 = file_inode(file2);
4304 rc = ll_swap_layouts_close(och, inode, inode2);
4306 rc = ll_swap_layouts(file, file2, &lsl);
4312 case LL_IOC_LOV_GETSTRIPE:
4313 case LL_IOC_LOV_GETSTRIPE_NEW:
4314 RETURN(ll_file_getstripe(inode, uarg, 0));
4315 case LL_IOC_GROUP_LOCK:
4316 RETURN(ll_get_grouplock(inode, file, arg));
4317 case LL_IOC_GROUP_UNLOCK:
4318 RETURN(ll_put_grouplock(inode, file, arg));
4319 case LL_IOC_DATA_VERSION: {
4320 struct ioc_data_version idv;
4323 if (copy_from_user(&idv, uarg, sizeof(idv)))
4326 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
4327 rc = ll_ioc_data_version(inode, &idv);
4329 if (rc == 0 && copy_to_user(uarg, &idv, sizeof(idv)))
4334 case LL_IOC_HSM_STATE_GET: {
4335 struct md_op_data *op_data;
4336 struct hsm_user_state *hus;
4339 if (!ll_access_ok(uarg, sizeof(*hus)))
4346 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4347 LUSTRE_OPC_ANY, hus);
4348 if (IS_ERR(op_data)) {
4349 rc = PTR_ERR(op_data);
4351 rc = obd_iocontrol(cmd, ll_i2mdexp(inode),
4352 sizeof(*op_data), op_data, NULL);
4354 if (copy_to_user(uarg, hus, sizeof(*hus)))
4357 ll_finish_md_op_data(op_data);
4362 case LL_IOC_HSM_STATE_SET: {
4363 struct hsm_state_set *hss;
4370 if (copy_from_user(hss, uarg, sizeof(*hss)))
4373 rc = ll_hsm_state_set(inode, hss);
4378 case LL_IOC_HSM_ACTION: {
4379 struct md_op_data *op_data;
4380 struct hsm_current_action *hca;
4384 if (!ll_access_ok(uarg, sizeof(*hca)))
4391 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4392 LUSTRE_OPC_ANY, hca);
4393 if (IS_ERR(op_data)) {
4395 RETURN(PTR_ERR(op_data));
4398 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4401 GOTO(skip_copy, rc);
4403 /* The hsm_current_action retreived from the server could
4404 * contain corrupt information. If it is incorrect data collect
4405 * debug information. We still send the data even if incorrect
4406 * to user land to handle.
4408 action = hsm_user_action2name(hca->hca_action);
4409 if (strcmp(action, "UNKNOWN") == 0 ||
4410 hca->hca_state > HPS_DONE) {
4412 "HSM current state %s action %s, offset = %llu, length %llu\n",
4413 hsm_progress_state2name(hca->hca_state), action,
4414 hca->hca_location.offset, hca->hca_location.length);
4417 if (copy_to_user(uarg, hca, sizeof(*hca)))
4420 ll_finish_md_op_data(op_data);
4424 case LL_IOC_SET_LEASE_OLD: {
4425 struct ll_ioc_lease ioc = { .lil_mode = arg };
4427 RETURN(ll_file_set_lease(file, &ioc, 0));
4429 case LL_IOC_SET_LEASE: {
4430 struct ll_ioc_lease ioc;
4432 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
4435 RETURN(ll_file_set_lease(file, &ioc, uarg));
4437 case LL_IOC_GET_LEASE: {
4438 struct ll_inode_info *lli = ll_i2info(inode);
4439 struct ldlm_lock *lock = NULL;
4442 mutex_lock(&lli->lli_och_mutex);
4443 if (fd->fd_lease_och != NULL) {
4444 struct obd_client_handle *och = fd->fd_lease_och;
4446 lock = ldlm_handle2lock(&och->och_lease_handle);
4448 lock_res_and_lock(lock);
4449 if (!ldlm_is_cancel(lock))
4450 fmode = och->och_flags;
4452 unlock_res_and_lock(lock);
4453 LDLM_LOCK_PUT(lock);
4456 mutex_unlock(&lli->lli_och_mutex);
4458 RETURN(ll_lease_type_from_fmode(fmode));
4460 case LL_IOC_HSM_IMPORT: {
4461 struct hsm_user_import *hui;
4467 if (copy_from_user(hui, uarg, sizeof(*hui)))
4470 rc = ll_hsm_import(inode, file, hui);
4475 case LL_IOC_FUTIMES_3: {
4476 struct ll_futimes_3 lfu;
4478 if (copy_from_user(&lfu, uarg, sizeof(lfu)))
4481 RETURN(ll_file_futimes_3(file, &lfu));
4483 case LL_IOC_LADVISE: {
4484 struct llapi_ladvise_hdr *k_ladvise_hdr;
4485 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4488 int alloc_size = sizeof(*k_ladvise_hdr);
4491 u_ladvise_hdr = uarg;
4492 OBD_ALLOC_PTR(k_ladvise_hdr);
4493 if (k_ladvise_hdr == NULL)
4496 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4497 GOTO(out_ladvise, rc = -EFAULT);
4499 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4500 k_ladvise_hdr->lah_count < 1)
4501 GOTO(out_ladvise, rc = -EINVAL);
4503 num_advise = k_ladvise_hdr->lah_count;
4504 if (num_advise >= LAH_COUNT_MAX)
4505 GOTO(out_ladvise, rc = -EFBIG);
4507 OBD_FREE_PTR(k_ladvise_hdr);
4508 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4509 lah_advise[num_advise]);
4510 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4511 if (k_ladvise_hdr == NULL)
4515 * TODO: submit multiple advices to one server in a single RPC
4517 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4518 GOTO(out_ladvise, rc = -EFAULT);
4520 for (i = 0; i < num_advise; i++) {
4521 struct llapi_lu_ladvise *k_ladvise =
4522 &k_ladvise_hdr->lah_advise[i];
4523 struct llapi_lu_ladvise __user *u_ladvise =
4524 &u_ladvise_hdr->lah_advise[i];
4526 rc = ll_ladvise_sanity(inode, k_ladvise);
4528 GOTO(out_ladvise, rc);
4530 switch (k_ladvise->lla_advice) {
4531 case LU_LADVISE_LOCKNOEXPAND:
4532 rc = ll_lock_noexpand(file,
4533 k_ladvise->lla_peradvice_flags);
4534 GOTO(out_ladvise, rc);
4535 case LU_LADVISE_LOCKAHEAD:
4537 rc = ll_file_lock_ahead(file, k_ladvise);
4540 GOTO(out_ladvise, rc);
4543 &u_ladvise->lla_lockahead_result))
4544 GOTO(out_ladvise, rc = -EFAULT);
4547 rc = ll_ladvise(inode, file,
4548 k_ladvise_hdr->lah_flags,
4551 GOTO(out_ladvise, rc);
4558 OBD_FREE(k_ladvise_hdr, alloc_size);
4561 case LL_IOC_FLR_SET_MIRROR: {
4562 /* mirror I/O must be direct to avoid polluting page cache
4564 if (!(file->f_flags & O_DIRECT))
4567 fd->fd_designated_mirror = arg;
4570 case LL_IOC_HEAT_GET: {
4571 struct lu_heat uheat;
4572 struct lu_heat *heat;
4575 if (copy_from_user(&uheat, uarg, sizeof(uheat)))
4578 if (uheat.lh_count > OBD_HEAT_COUNT)
4579 uheat.lh_count = OBD_HEAT_COUNT;
4581 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4582 OBD_ALLOC(heat, size);
4586 heat->lh_count = uheat.lh_count;
4587 ll_heat_get(inode, heat);
4588 rc = copy_to_user(uarg, heat, size);
4589 OBD_FREE(heat, size);
4590 RETURN(rc ? -EFAULT : 0);
4592 case LL_IOC_HEAT_SET: {
4595 if (copy_from_user(&flags, uarg, sizeof(flags)))
4598 rc = ll_heat_set(inode, flags);
4601 case LL_IOC_PCC_DETACH: {
4602 struct lu_pcc_detach *detach;
4604 OBD_ALLOC_PTR(detach);
4608 if (copy_from_user(detach, uarg, sizeof(*detach)))
4609 GOTO(out_detach_free, rc = -EFAULT);
4611 if (!S_ISREG(inode->i_mode))
4612 GOTO(out_detach_free, rc = -EINVAL);
4614 if (!inode_owner_or_capable(&init_user_ns, inode))
4615 GOTO(out_detach_free, rc = -EPERM);
4617 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4619 OBD_FREE_PTR(detach);
4622 case LL_IOC_PCC_STATE: {
4623 struct lu_pcc_state __user *ustate = uarg;
4624 struct lu_pcc_state *state;
4626 OBD_ALLOC_PTR(state);
4630 if (copy_from_user(state, ustate, sizeof(*state)))
4631 GOTO(out_state, rc = -EFAULT);
4633 rc = pcc_ioctl_state(file, inode, state);
4635 GOTO(out_state, rc);
4637 if (copy_to_user(ustate, state, sizeof(*state)))
4638 GOTO(out_state, rc = -EFAULT);
4641 OBD_FREE_PTR(state);
4645 rc = ll_iocontrol(inode, file, cmd, uarg);
4648 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, uarg));
4652 loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4654 struct inode *inode = file_inode(file);
4657 struct cl_lseek_io *lsio;
4664 env = cl_env_get(&refcheck);
4666 RETURN(PTR_ERR(env));
4668 io = vvp_env_thread_io(env);
4669 io->ci_obj = ll_i2info(inode)->lli_clob;
4670 ll_io_set_mirror(io, file);
4672 lsio = &io->u.ci_lseek;
4673 lsio->ls_start = offset;
4674 lsio->ls_whence = whence;
4675 lsio->ls_result = -ENXIO;
4678 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4680 struct vvp_io *vio = vvp_env_io(env);
4682 vio->vui_fd = file->private_data;
4683 rc = cl_io_loop(env, io);
4687 retval = rc ? : lsio->ls_result;
4688 cl_io_fini(env, io);
4689 } while (unlikely(io->ci_need_restart));
4691 cl_env_put(env, &refcheck);
4693 /* Without the key, SEEK_HOLE return value has to be
4694 * rounded up to next LUSTRE_ENCRYPTION_UNIT_SIZE.
4696 if (llcrypt_require_key(inode) == -ENOKEY && whence == SEEK_HOLE)
4697 retval = round_up(retval, LUSTRE_ENCRYPTION_UNIT_SIZE);
4702 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4704 struct inode *inode = file_inode(file);
4705 loff_t retval = offset, eof = 0;
4706 ktime_t kstart = ktime_get();
4710 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4711 PFID(ll_inode2fid(inode)), inode, retval, retval,
4714 if (origin == SEEK_END) {
4715 retval = ll_glimpse_size(inode);
4718 eof = i_size_read(inode);
4721 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4725 /* flush local cache first if any */
4726 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4729 retval = ll_lseek(file, offset, origin);
4732 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4734 retval = generic_file_llseek_size(file, offset, origin,
4735 ll_file_maxbytes(inode), eof);
4738 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4739 ktime_us_delta(ktime_get(), kstart));
4743 static int ll_flush(struct file *file, fl_owner_t id)
4745 struct inode *inode = file_inode(file);
4746 struct ll_inode_info *lli = ll_i2info(inode);
4747 struct ll_file_data *fd = file->private_data;
4750 LASSERT(!S_ISDIR(inode->i_mode));
4752 /* catch async errors that were recorded back when async writeback
4753 * failed for pages in this mapping. */
4754 rc = lli->lli_async_rc;
4755 lli->lli_async_rc = 0;
4756 if (lli->lli_clob != NULL) {
4757 err = lov_read_and_clear_async_rc(lli->lli_clob);
4762 /* The application has been told write failure already.
4763 * Do not report failure again. */
4764 if (fd->fd_write_failed)
4766 return rc ? -EIO : 0;
4770 * Called to make sure a portion of file has been written out.
4771 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4773 * Return how many pages have been written.
4775 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4776 enum cl_fsync_mode mode, int ignore_layout)
4780 struct cl_fsync_io *fio;
4785 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4786 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4789 env = cl_env_get(&refcheck);
4791 RETURN(PTR_ERR(env));
4793 io = vvp_env_thread_io(env);
4794 io->ci_obj = ll_i2info(inode)->lli_clob;
4795 io->ci_ignore_layout = ignore_layout;
4797 /* initialize parameters for sync */
4798 fio = &io->u.ci_fsync;
4799 fio->fi_start = start;
4801 fio->fi_fid = ll_inode2fid(inode);
4802 fio->fi_mode = mode;
4803 fio->fi_nr_written = 0;
4805 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4806 result = cl_io_loop(env, io);
4808 result = io->ci_result;
4810 result = fio->fi_nr_written;
4811 cl_io_fini(env, io);
4812 cl_env_put(env, &refcheck);
4818 * When dentry is provided (the 'else' case), file_dentry() may be
4819 * null and dentry must be used directly rather than pulled from
4820 * file_dentry() as is done otherwise.
4823 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4825 struct dentry *dentry = file_dentry(file);
4826 struct inode *inode = dentry->d_inode;
4827 struct ll_inode_info *lli = ll_i2info(inode);
4828 struct ptlrpc_request *req;
4829 ktime_t kstart = ktime_get();
4835 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4836 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4838 /* fsync's caller has already called _fdata{sync,write}, we want
4839 * that IO to finish before calling the osc and mdc sync methods */
4840 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4842 /* catch async errors that were recorded back when async writeback
4843 * failed for pages in this mapping. */
4844 if (!S_ISDIR(inode->i_mode)) {
4845 err = lli->lli_async_rc;
4846 lli->lli_async_rc = 0;
4849 if (lli->lli_clob != NULL) {
4850 err = lov_read_and_clear_async_rc(lli->lli_clob);
4856 if (S_ISREG(inode->i_mode) && !lli->lli_synced_to_mds) {
4858 * only the first sync on MDS makes sense,
4859 * everything else is stored on OSTs
4861 err = md_fsync(ll_i2sbi(inode)->ll_md_exp,
4862 ll_inode2fid(inode), &req);
4866 lli->lli_synced_to_mds = true;
4867 ptlrpc_req_finished(req);
4871 if (S_ISREG(inode->i_mode)) {
4872 struct ll_file_data *fd = file->private_data;
4875 /* Sync metadata on MDT first, and then sync the cached data
4878 err = pcc_fsync(file, start, end, datasync, &cached);
4880 err = cl_sync_file_range(inode, start, end,
4882 if (rc == 0 && err < 0)
4885 fd->fd_write_failed = true;
4887 fd->fd_write_failed = false;
4891 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4892 ktime_us_delta(ktime_get(), kstart));
4897 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4899 struct inode *inode = file_inode(file);
4900 struct ll_sb_info *sbi = ll_i2sbi(inode);
4901 struct ldlm_enqueue_info einfo = {
4902 .ei_type = LDLM_FLOCK,
4903 .ei_cb_cp = ldlm_flock_completion_ast,
4904 .ei_cbdata = file_lock,
4906 struct md_op_data *op_data;
4907 struct lustre_handle lockh = { 0 };
4908 union ldlm_policy_data flock = { { 0 } };
4909 int fl_type = file_lock->fl_type;
4910 ktime_t kstart = ktime_get();
4916 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4917 PFID(ll_inode2fid(inode)), file_lock);
4919 if (file_lock->fl_flags & FL_FLOCK) {
4920 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4921 /* flocks are whole-file locks */
4922 flock.l_flock.end = OFFSET_MAX;
4923 /* For flocks owner is determined by the local file desctiptor*/
4924 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4925 } else if (file_lock->fl_flags & FL_POSIX) {
4926 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4927 flock.l_flock.start = file_lock->fl_start;
4928 flock.l_flock.end = file_lock->fl_end;
4932 flock.l_flock.pid = file_lock->fl_pid;
4934 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4935 /* Somewhat ugly workaround for svc lockd.
4936 * lockd installs custom fl_lmops->lm_compare_owner that checks
4937 * for the fl_owner to be the same (which it always is on local node
4938 * I guess between lockd processes) and then compares pid.
4939 * As such we assign pid to the owner field to make it all work,
4940 * conflict with normal locks is unlikely since pid space and
4941 * pointer space for current->files are not intersecting */
4942 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4943 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4948 einfo.ei_mode = LCK_PR;
4951 /* An unlock request may or may not have any relation to
4952 * existing locks so we may not be able to pass a lock handle
4953 * via a normal ldlm_lock_cancel() request. The request may even
4954 * unlock a byte range in the middle of an existing lock. In
4955 * order to process an unlock request we need all of the same
4956 * information that is given with a normal read or write record
4957 * lock request. To avoid creating another ldlm unlock (cancel)
4958 * message we'll treat a LCK_NL flock request as an unlock. */
4959 einfo.ei_mode = LCK_NL;
4962 einfo.ei_mode = LCK_PW;
4966 CERROR("%s: fcntl from '%s' unknown lock type=%d: rc = %d\n",
4967 sbi->ll_fsname, current->comm, fl_type, rc);
4982 flags = LDLM_FL_BLOCK_NOWAIT;
4988 flags = LDLM_FL_TEST_LOCK;
4992 CERROR("%s: fcntl from '%s' unknown lock command=%d: rc = %d\n",
4993 sbi->ll_fsname, current->comm, cmd, rc);
4997 /* Save the old mode so that if the mode in the lock changes we
4998 * can decrement the appropriate reader or writer refcount. */
4999 file_lock->fl_type = einfo.ei_mode;
5001 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
5002 LUSTRE_OPC_ANY, NULL);
5003 if (IS_ERR(op_data))
5004 RETURN(PTR_ERR(op_data));
5006 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
5007 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
5008 flock.l_flock.pid, flags, einfo.ei_mode,
5009 flock.l_flock.start, flock.l_flock.end);
5011 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
5014 /* Restore the file lock type if not TEST lock. */
5015 if (!(flags & LDLM_FL_TEST_LOCK))
5016 file_lock->fl_type = fl_type;
5018 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
5019 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
5020 !(flags & LDLM_FL_TEST_LOCK))
5021 rc2 = locks_lock_file_wait(file, file_lock);
5023 if ((file_lock->fl_flags & FL_FLOCK) &&
5024 (rc == 0 || file_lock->fl_type == F_UNLCK))
5025 rc2 = flock_lock_file_wait(file, file_lock);
5026 if ((file_lock->fl_flags & FL_POSIX) &&
5027 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
5028 !(flags & LDLM_FL_TEST_LOCK))
5029 rc2 = posix_lock_file_wait(file, file_lock);
5030 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
5032 if (rc2 && file_lock->fl_type != F_UNLCK) {
5033 einfo.ei_mode = LCK_NL;
5034 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
5039 ll_finish_md_op_data(op_data);
5042 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
5043 ktime_us_delta(ktime_get(), kstart));
5047 int ll_get_fid_by_name(struct inode *parent, const char *name,
5048 int namelen, struct lu_fid *fid,
5049 struct inode **inode)
5051 struct md_op_data *op_data = NULL;
5052 struct mdt_body *body;
5053 struct ptlrpc_request *req;
5057 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
5058 LUSTRE_OPC_ANY, NULL);
5059 if (IS_ERR(op_data))
5060 RETURN(PTR_ERR(op_data));
5062 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
5063 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
5064 ll_finish_md_op_data(op_data);
5068 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
5070 GOTO(out_req, rc = -EFAULT);
5072 *fid = body->mbo_fid1;
5075 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
5077 ptlrpc_req_finished(req);
5081 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
5082 const char *name, __u32 flags)
5084 struct dentry *dchild = NULL;
5085 struct inode *child_inode = NULL;
5086 struct md_op_data *op_data;
5087 struct ptlrpc_request *request = NULL;
5088 struct obd_client_handle *och = NULL;
5090 struct mdt_body *body;
5091 __u64 data_version = 0;
5092 size_t namelen = strlen(name);
5093 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
5097 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
5098 PFID(ll_inode2fid(parent)), name,
5099 lum->lum_stripe_offset, lum->lum_stripe_count);
5101 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
5102 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
5103 lustre_swab_lmv_user_md(lum);
5105 /* Get child FID first */
5106 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
5109 dchild = d_lookup(file_dentry(file), &qstr);
5111 if (dchild->d_inode)
5112 child_inode = igrab(dchild->d_inode);
5117 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
5126 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
5127 OBD_CONNECT2_DIR_MIGRATE)) {
5128 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
5129 ll_dir_striped(child_inode)) {
5130 CERROR("%s: MDT doesn't support stripe directory "
5131 "migration!\n", ll_i2sbi(parent)->ll_fsname);
5132 GOTO(out_iput, rc = -EOPNOTSUPP);
5137 * lfs migrate command needs to be blocked on the client
5138 * by checking the migrate FID against the FID of the
5141 if (is_root_inode(child_inode))
5142 GOTO(out_iput, rc = -EINVAL);
5144 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
5145 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
5146 if (IS_ERR(op_data))
5147 GOTO(out_iput, rc = PTR_ERR(op_data));
5149 inode_lock(child_inode);
5150 op_data->op_fid3 = *ll_inode2fid(child_inode);
5151 if (!fid_is_sane(&op_data->op_fid3)) {
5152 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
5153 ll_i2sbi(parent)->ll_fsname, name,
5154 PFID(&op_data->op_fid3));
5155 GOTO(out_unlock, rc = -EINVAL);
5158 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
5159 op_data->op_data = lum;
5160 op_data->op_data_size = lumlen;
5162 /* migrate dirent only for subdirs if MDS_MIGRATE_NSONLY set */
5163 if (S_ISDIR(child_inode->i_mode) && (flags & MDS_MIGRATE_NSONLY) &&
5164 lmv_dir_layout_changing(ll_i2info(parent)->lli_lsm_md))
5165 op_data->op_bias |= MDS_MIGRATE_NSONLY;
5168 if (S_ISREG(child_inode->i_mode)) {
5169 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
5173 GOTO(out_unlock, rc);
5176 rc = ll_data_version(child_inode, &data_version,
5179 GOTO(out_close, rc);
5181 op_data->op_open_handle = och->och_open_handle;
5182 op_data->op_data_version = data_version;
5183 op_data->op_lease_handle = och->och_lease_handle;
5184 op_data->op_bias |= MDS_CLOSE_MIGRATE;
5186 spin_lock(&och->och_mod->mod_open_req->rq_lock);
5187 och->och_mod->mod_open_req->rq_replay = 0;
5188 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
5191 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
5192 op_data->op_name, op_data->op_namelen,
5193 op_data->op_name, op_data->op_namelen, &request);
5195 LASSERT(request != NULL);
5196 ll_update_times(request, parent);
5199 if (rc == 0 || rc == -EAGAIN) {
5200 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
5201 LASSERT(body != NULL);
5203 /* If the server does release layout lock, then we cleanup
5204 * the client och here, otherwise release it in out_close: */
5205 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
5206 obd_mod_put(och->och_mod);
5207 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
5209 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
5215 if (request != NULL) {
5216 ptlrpc_req_finished(request);
5220 /* Try again if the lease has cancelled. */
5221 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
5226 ll_lease_close(och, child_inode, NULL);
5228 clear_nlink(child_inode);
5230 inode_unlock(child_inode);
5231 ll_finish_md_op_data(op_data);
5238 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
5240 struct ll_file_data *fd = file->private_data;
5244 * In order to avoid flood of warning messages, only print one message
5245 * for one file. And the entire message rate on the client is limited
5246 * by CDEBUG_LIMIT too.
5248 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
5249 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
5250 CDEBUG_LIMIT(D_CONSOLE,
5251 "flock disabled, mount with '-o [local]flock' to enable\r\n");
5257 * test if some locks matching bits and l_req_mode are acquired
5258 * - bits can be in different locks
5259 * - if found clear the common lock bits in *bits
5260 * - the bits not found, are kept in *bits
5262 * \param bits [IN] searched lock bits [IN]
5263 * \param l_req_mode [IN] searched lock mode
5264 * \retval boolean, true iff all bits are found
5266 int ll_have_md_lock(struct obd_export *exp, struct inode *inode, __u64 *bits,
5267 enum ldlm_mode l_req_mode)
5269 struct lustre_handle lockh;
5270 union ldlm_policy_data policy;
5271 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
5272 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5281 fid = &ll_i2info(inode)->lli_fid;
5282 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5283 ldlm_lockname[mode]);
5285 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5286 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5287 policy.l_inodebits.bits = *bits & BIT(i);
5288 if (policy.l_inodebits.bits == 0)
5291 if (md_lock_match(exp, flags, fid, LDLM_IBITS, &policy, mode,
5293 struct ldlm_lock *lock;
5295 lock = ldlm_handle2lock(&lockh);
5298 ~(lock->l_policy_data.l_inodebits.bits);
5299 LDLM_LOCK_PUT(lock);
5301 *bits &= ~policy.l_inodebits.bits;
5308 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5309 struct lustre_handle *lockh, __u64 flags,
5310 enum ldlm_mode mode)
5312 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5317 fid = &ll_i2info(inode)->lli_fid;
5318 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5320 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5321 fid, LDLM_IBITS, &policy, mode, lockh);
5326 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5328 /* Already unlinked. Just update nlink and return success */
5329 if (rc == -ENOENT) {
5331 /* If it is striped directory, and there is bad stripe
5332 * Let's revalidate the dentry again, instead of returning
5334 if (ll_dir_striped(inode))
5337 /* This path cannot be hit for regular files unless in
5338 * case of obscure races, so no need to to validate
5340 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5342 } else if (rc != 0) {
5343 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5344 "%s: revalidate FID "DFID" error: rc = %d\n",
5345 ll_i2sbi(inode)->ll_fsname,
5346 PFID(ll_inode2fid(inode)), rc);
5352 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5354 struct inode *parent;
5355 struct inode *inode = dentry->d_inode;
5356 struct obd_export *exp = ll_i2mdexp(inode);
5357 struct lookup_intent oit = {
5360 struct ptlrpc_request *req = NULL;
5361 struct md_op_data *op_data;
5362 const char *name = NULL;
5367 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5368 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5370 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5371 parent = dentry->d_parent->d_inode;
5372 name = dentry->d_name.name;
5373 namelen = dentry->d_name.len;
5378 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5379 LUSTRE_OPC_ANY, NULL);
5380 if (IS_ERR(op_data))
5381 RETURN(PTR_ERR(op_data));
5383 /* Call getattr by fid */
5384 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5385 op_data->op_flags = MF_GETATTR_BY_FID;
5386 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5387 ll_finish_md_op_data(op_data);
5389 rc = ll_inode_revalidate_fini(inode, rc);
5393 rc = ll_revalidate_it_finish(req, &oit, dentry);
5395 ll_intent_release(&oit);
5399 /* Unlinked? Unhash dentry, so it is not picked up later by
5400 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5401 * here to preserve get_cwd functionality on 2.6.
5403 if (!dentry->d_inode->i_nlink)
5404 d_lustre_invalidate(dentry);
5406 ll_lookup_finish_locks(&oit, dentry);
5408 ptlrpc_req_finished(req);
5413 static int ll_merge_md_attr(struct inode *inode)
5415 struct ll_inode_info *lli = ll_i2info(inode);
5416 struct cl_attr attr = { 0 };
5419 if (!lli->lli_lsm_md)
5422 down_read(&lli->lli_lsm_sem);
5423 if (!lmv_dir_striped(lli->lli_lsm_md)) {
5424 up_read(&lli->lli_lsm_sem);
5427 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
5428 &attr, ll_md_blocking_ast);
5429 up_read(&lli->lli_lsm_sem);
5433 spin_lock(&inode->i_lock);
5434 set_nlink(inode, attr.cat_nlink);
5435 spin_unlock(&inode->i_lock);
5437 inode->i_blocks = attr.cat_blocks;
5438 i_size_write(inode, attr.cat_size);
5440 ll_i2info(inode)->lli_atime = attr.cat_atime;
5441 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5442 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5447 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5448 unsigned int flags, bool foreign)
5450 struct inode *inode = de->d_inode;
5451 struct ll_sb_info *sbi = ll_i2sbi(inode);
5452 struct ll_inode_info *lli = ll_i2info(inode);
5453 struct inode *dir = de->d_parent->d_inode;
5454 bool need_glimpse = true;
5455 ktime_t kstart = ktime_get();
5458 /* The OST object(s) determine the file size, blocks and mtime. */
5459 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5460 request_mask & STATX_MTIME))
5461 need_glimpse = false;
5463 if (dentry_may_statahead(dir, de))
5464 ll_start_statahead(dir, de, need_glimpse &&
5465 !(flags & AT_STATX_DONT_SYNC));
5467 if (flags & AT_STATX_DONT_SYNC)
5468 GOTO(fill_attr, rc = 0);
5470 rc = ll_inode_revalidate(de, IT_GETATTR);
5474 /* foreign file/dir are always of zero length, so don't
5475 * need to validate size.
5477 if (S_ISREG(inode->i_mode) && !foreign) {
5481 GOTO(fill_attr, rc);
5483 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5484 if (cached && rc < 0)
5488 GOTO(fill_attr, rc);
5491 * If the returned attr is masked with OBD_MD_FLSIZE &
5492 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5493 * or blocks obtained from MDT is strictly correct, and the file
5494 * is usually not being modified by clients, and the [a|m|c]time
5495 * got from MDT is also strictly correct.
5496 * Under this circumstance, it does not need to send glimpse
5497 * RPCs to OSTs for file attributes such as the size and blocks.
5499 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5500 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5501 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5502 inode->i_mtime.tv_sec = lli->lli_mtime;
5503 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5504 inode->i_atime.tv_sec = lli->lli_atime;
5505 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5506 inode->i_ctime.tv_sec = lli->lli_ctime;
5507 GOTO(fill_attr, rc);
5510 /* In case of restore, the MDT has the right size and has
5511 * already send it back without granting the layout lock,
5512 * inode is up-to-date so glimpse is useless.
5513 * Also to glimpse we need the layout, in case of a running
5514 * restore the MDT holds the layout lock so the glimpse will
5515 * block up to the end of restore (getattr will block)
5517 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5518 rc = ll_glimpse_size(inode);
5523 /* If object isn't regular a file then don't validate size. */
5524 /* foreign dir is not striped dir */
5525 if (ll_dir_striped(inode) && !foreign) {
5526 rc = ll_merge_md_attr(inode);
5531 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5532 inode->i_atime.tv_sec = lli->lli_atime;
5533 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5534 inode->i_mtime.tv_sec = lli->lli_mtime;
5535 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5536 inode->i_ctime.tv_sec = lli->lli_ctime;
5540 CFS_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5542 if (ll_need_32bit_api(sbi)) {
5543 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5544 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5545 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5547 stat->ino = inode->i_ino;
5548 stat->dev = inode->i_sb->s_dev;
5549 stat->rdev = inode->i_rdev;
5552 /* foreign symlink to be exposed as a real symlink */
5554 stat->mode = inode->i_mode;
5556 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5558 stat->uid = inode->i_uid;
5559 stat->gid = inode->i_gid;
5560 stat->atime = inode->i_atime;
5561 stat->mtime = inode->i_mtime;
5562 stat->ctime = inode->i_ctime;
5563 /* stat->blksize is used to tell about preferred IO size */
5564 if (sbi->ll_stat_blksize)
5565 stat->blksize = sbi->ll_stat_blksize;
5566 else if (S_ISREG(inode->i_mode))
5567 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
5568 LL_MAX_BLKSIZE_BITS);
5570 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5572 stat->nlink = inode->i_nlink;
5573 stat->size = i_size_read(inode);
5574 stat->blocks = inode->i_blocks;
5576 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5577 if (flags & AT_STATX_DONT_SYNC) {
5578 if (stat->size == 0 &&
5579 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5580 stat->size = lli->lli_lazysize;
5581 if (stat->blocks == 0 &&
5582 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5583 stat->blocks = lli->lli_lazyblocks;
5586 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5587 stat->result_mask |= STATX_BTIME;
5588 stat->btime.tv_sec = lli->lli_btime;
5591 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5592 #ifdef HAVE_LUSTRE_CRYPTO
5593 stat->attributes_mask |= STATX_ATTR_ENCRYPTED;
5595 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5596 /* if Lustre specific LUSTRE_ENCRYPT_FL flag is set, also set
5597 * ext4 equivalent to please statx
5599 if (stat->attributes & LUSTRE_ENCRYPT_FL)
5600 stat->attributes |= STATX_ATTR_ENCRYPTED;
5601 stat->result_mask &= request_mask;
5604 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5605 ktime_us_delta(ktime_get(), kstart));
5610 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5611 int ll_getattr(struct user_namespace *mnt_userns, const struct path *path,
5612 struct kstat *stat, u32 request_mask, unsigned int flags)
5614 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5618 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5620 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5621 AT_STATX_SYNC_AS_STAT, false);
5625 int cl_falloc(struct file *file, struct inode *inode, int mode, loff_t offset,
5628 loff_t size = i_size_read(inode);
5636 env = cl_env_get(&refcheck);
5638 RETURN(PTR_ERR(env));
5640 io = vvp_env_thread_io(env);
5641 io->ci_obj = ll_i2info(inode)->lli_clob;
5642 ll_io_set_mirror(io, file);
5644 io->ci_verify_layout = 1;
5645 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5646 io->u.ci_setattr.sa_falloc_mode = mode;
5647 io->u.ci_setattr.sa_falloc_offset = offset;
5648 io->u.ci_setattr.sa_falloc_end = offset + len;
5649 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5651 CDEBUG(D_INODE, "UID %u GID %u PRJID %u\n",
5652 from_kuid(&init_user_ns, inode->i_uid),
5653 from_kgid(&init_user_ns, inode->i_gid),
5654 ll_i2info(inode)->lli_projid);
5656 io->u.ci_setattr.sa_falloc_uid = from_kuid(&init_user_ns, inode->i_uid);
5657 io->u.ci_setattr.sa_falloc_gid = from_kgid(&init_user_ns, inode->i_gid);
5658 io->u.ci_setattr.sa_falloc_projid = ll_i2info(inode)->lli_projid;
5660 if (io->u.ci_setattr.sa_falloc_end > size) {
5661 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5663 /* Check new size against VFS/VM file size limit and rlimit */
5664 rc = inode_newsize_ok(inode, newsize);
5667 if (newsize > ll_file_maxbytes(inode)) {
5668 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5669 (unsigned long long)newsize,
5670 ll_file_maxbytes(inode));
5677 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5679 rc = cl_io_loop(env, io);
5682 cl_io_fini(env, io);
5683 } while (unlikely(io->ci_need_restart));
5686 cl_env_put(env, &refcheck);
5690 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5692 struct inode *inode = file_inode(filp);
5695 if (offset < 0 || len <= 0)
5698 * Encrypted inodes can't handle collapse range or zero range or insert
5699 * range since we would need to re-encrypt blocks with a different IV or
5700 * XTS tweak (which are based on the logical block number).
5701 * Similar to what ext4 does.
5703 if (IS_ENCRYPTED(inode) &&
5704 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5705 FALLOC_FL_ZERO_RANGE)))
5706 RETURN(-EOPNOTSUPP);
5709 * mode == 0 (which is standard prealloc) and PUNCH is supported
5710 * Rest of mode options are not supported yet.
5712 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5713 RETURN(-EOPNOTSUPP);
5715 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5717 rc = cl_falloc(filp, inode, mode, offset, len);
5719 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5720 * used by Lustre in several places. Retuning it here would
5721 * confuse applications that explicity test for EOPNOTSUPP
5722 * (95) and fall back to ftruncate().
5724 if (rc == -ENOTSUPP)
5730 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5731 __u64 start, __u64 len)
5735 struct fiemap *fiemap;
5736 unsigned int extent_count = fieinfo->fi_extents_max;
5738 num_bytes = sizeof(*fiemap) + (extent_count *
5739 sizeof(struct fiemap_extent));
5740 OBD_ALLOC_LARGE(fiemap, num_bytes);
5745 fiemap->fm_flags = fieinfo->fi_flags;
5746 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5747 fiemap->fm_start = start;
5748 fiemap->fm_length = len;
5749 if (extent_count > 0 &&
5750 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5751 sizeof(struct fiemap_extent)) != 0)
5752 GOTO(out, rc = -EFAULT);
5754 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5756 if (IS_ENCRYPTED(inode)) {
5759 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5760 fiemap->fm_extents[i].fe_flags |=
5761 FIEMAP_EXTENT_DATA_ENCRYPTED |
5762 FIEMAP_EXTENT_ENCODED;
5765 fieinfo->fi_flags = fiemap->fm_flags;
5766 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5767 if (extent_count > 0 &&
5768 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5769 fiemap->fm_mapped_extents *
5770 sizeof(struct fiemap_extent)) != 0)
5771 GOTO(out, rc = -EFAULT);
5773 OBD_FREE_LARGE(fiemap, num_bytes);
5777 int ll_inode_permission(struct user_namespace *mnt_userns, struct inode *inode,
5781 struct ll_sb_info *sbi;
5782 struct root_squash_info *squash;
5783 struct cred *cred = NULL;
5784 const struct cred *old_cred = NULL;
5785 bool squash_id = false;
5786 ktime_t kstart = ktime_get();
5790 if (mask & MAY_NOT_BLOCK)
5794 * as root inode are NOT getting validated in lookup operation,
5795 * need to revalidate PERM before permission check.
5797 if (is_root_inode(inode)) {
5798 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_GETATTR);
5803 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5804 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5806 /* squash fsuid/fsgid if needed */
5807 sbi = ll_i2sbi(inode);
5808 squash = &sbi->ll_squash;
5809 if (unlikely(squash->rsi_uid != 0 &&
5810 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5811 !test_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags))) {
5815 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5816 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5817 squash->rsi_uid, squash->rsi_gid);
5819 /* update current process's credentials
5820 * and FS capability */
5821 cred = prepare_creds();
5825 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5826 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5827 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5828 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5830 old_cred = override_creds(cred);
5833 rc = generic_permission(mnt_userns, inode, mask);
5834 /* restore current process's credentials and FS capability */
5836 revert_creds(old_cred);
5841 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5842 ktime_us_delta(ktime_get(), kstart));
5847 /* -o localflock - only provides locally consistent flock locks */
5848 static const struct file_operations ll_file_operations = {
5849 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5850 # ifdef HAVE_SYNC_READ_WRITE
5851 .read = new_sync_read,
5852 .write = new_sync_write,
5854 .read_iter = ll_file_read_iter,
5855 .write_iter = ll_file_write_iter,
5856 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5857 .read = ll_file_read,
5858 .aio_read = ll_file_aio_read,
5859 .write = ll_file_write,
5860 .aio_write = ll_file_aio_write,
5861 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5862 .unlocked_ioctl = ll_file_ioctl,
5863 .open = ll_file_open,
5864 .release = ll_file_release,
5865 .mmap = ll_file_mmap,
5866 .llseek = ll_file_seek,
5867 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5868 .splice_read = generic_file_splice_read,
5870 .splice_read = pcc_file_splice_read,
5872 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5873 .splice_write = iter_file_splice_write,
5877 .fallocate = ll_fallocate,
5880 static const struct file_operations ll_file_operations_flock = {
5881 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5882 # ifdef HAVE_SYNC_READ_WRITE
5883 .read = new_sync_read,
5884 .write = new_sync_write,
5885 # endif /* HAVE_SYNC_READ_WRITE */
5886 .read_iter = ll_file_read_iter,
5887 .write_iter = ll_file_write_iter,
5888 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5889 .read = ll_file_read,
5890 .aio_read = ll_file_aio_read,
5891 .write = ll_file_write,
5892 .aio_write = ll_file_aio_write,
5893 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5894 .unlocked_ioctl = ll_file_ioctl,
5895 .open = ll_file_open,
5896 .release = ll_file_release,
5897 .mmap = ll_file_mmap,
5898 .llseek = ll_file_seek,
5899 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5900 .splice_read = generic_file_splice_read,
5902 .splice_read = pcc_file_splice_read,
5904 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5905 .splice_write = iter_file_splice_write,
5909 .flock = ll_file_flock,
5910 .lock = ll_file_flock,
5911 .fallocate = ll_fallocate,
5914 /* These are for -o noflock - to return ENOSYS on flock calls */
5915 static const struct file_operations ll_file_operations_noflock = {
5916 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5917 # ifdef HAVE_SYNC_READ_WRITE
5918 .read = new_sync_read,
5919 .write = new_sync_write,
5920 # endif /* HAVE_SYNC_READ_WRITE */
5921 .read_iter = ll_file_read_iter,
5922 .write_iter = ll_file_write_iter,
5923 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5924 .read = ll_file_read,
5925 .aio_read = ll_file_aio_read,
5926 .write = ll_file_write,
5927 .aio_write = ll_file_aio_write,
5928 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5929 .unlocked_ioctl = ll_file_ioctl,
5930 .open = ll_file_open,
5931 .release = ll_file_release,
5932 .mmap = ll_file_mmap,
5933 .llseek = ll_file_seek,
5934 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5935 .splice_read = generic_file_splice_read,
5937 .splice_read = pcc_file_splice_read,
5939 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5940 .splice_write = iter_file_splice_write,
5944 .flock = ll_file_noflock,
5945 .lock = ll_file_noflock,
5946 .fallocate = ll_fallocate,
5949 const struct inode_operations ll_file_inode_operations = {
5950 .setattr = ll_setattr,
5951 .getattr = ll_getattr,
5952 .permission = ll_inode_permission,
5953 #ifdef HAVE_IOP_XATTR
5954 .setxattr = ll_setxattr,
5955 .getxattr = ll_getxattr,
5956 .removexattr = ll_removexattr,
5958 .listxattr = ll_listxattr,
5959 .fiemap = ll_fiemap,
5960 .get_acl = ll_get_acl,
5961 #ifdef HAVE_IOP_SET_ACL
5962 .set_acl = ll_set_acl,
5966 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
5968 const struct file_operations *fops = &ll_file_operations_noflock;
5970 if (test_bit(LL_SBI_FLOCK, sbi->ll_flags))
5971 fops = &ll_file_operations_flock;
5972 else if (test_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags))
5973 fops = &ll_file_operations;
5978 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5980 struct ll_inode_info *lli = ll_i2info(inode);
5981 struct cl_object *obj = lli->lli_clob;
5990 env = cl_env_get(&refcheck);
5992 RETURN(PTR_ERR(env));
5994 rc = cl_conf_set(env, lli->lli_clob, conf);
5998 if (conf->coc_opc == OBJECT_CONF_SET) {
5999 struct ldlm_lock *lock = conf->coc_lock;
6000 struct cl_layout cl = {
6004 LASSERT(lock != NULL);
6005 LASSERT(ldlm_has_layout(lock));
6007 /* it can only be allowed to match after layout is
6008 * applied to inode otherwise false layout would be
6009 * seen. Applying layout shoud happen before dropping
6010 * the intent lock. */
6011 ldlm_lock_allow_match(lock);
6013 rc = cl_object_layout_get(env, obj, &cl);
6018 DFID": layout version change: %u -> %u\n",
6019 PFID(&lli->lli_fid), ll_layout_version_get(lli),
6021 ll_layout_version_set(lli, cl.cl_layout_gen);
6025 cl_env_put(env, &refcheck);
6027 RETURN(rc < 0 ? rc : 0);
6030 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
6031 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
6034 struct ll_sb_info *sbi = ll_i2sbi(inode);
6035 struct ptlrpc_request *req;
6042 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
6043 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
6044 lock->l_lvb_data, lock->l_lvb_len);
6046 if (lock->l_lvb_data != NULL)
6049 /* if layout lock was granted right away, the layout is returned
6050 * within DLM_LVB of dlm reply; otherwise if the lock was ever
6051 * blocked and then granted via completion ast, we have to fetch
6052 * layout here. Please note that we can't use the LVB buffer in
6053 * completion AST because it doesn't have a large enough buffer */
6054 rc = ll_get_default_mdsize(sbi, &lmmsize);
6058 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
6059 XATTR_NAME_LOV, lmmsize, &req);
6062 GOTO(out, rc = 0); /* empty layout */
6069 if (lmmsize == 0) /* empty layout */
6072 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
6074 GOTO(out, rc = -EFAULT);
6076 OBD_ALLOC_LARGE(lvbdata, lmmsize);
6077 if (lvbdata == NULL)
6078 GOTO(out, rc = -ENOMEM);
6080 memcpy(lvbdata, lmm, lmmsize);
6081 lock_res_and_lock(lock);
6082 if (unlikely(lock->l_lvb_data == NULL)) {
6083 lock->l_lvb_type = LVB_T_LAYOUT;
6084 lock->l_lvb_data = lvbdata;
6085 lock->l_lvb_len = lmmsize;
6088 unlock_res_and_lock(lock);
6091 OBD_FREE_LARGE(lvbdata, lmmsize);
6096 ptlrpc_req_finished(req);
6101 * Apply the layout to the inode. Layout lock is held and will be released
6104 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
6105 struct inode *inode)
6107 struct ll_inode_info *lli = ll_i2info(inode);
6108 struct ll_sb_info *sbi = ll_i2sbi(inode);
6109 struct ldlm_lock *lock;
6110 struct cl_object_conf conf;
6113 bool wait_layout = false;
6116 LASSERT(lustre_handle_is_used(lockh));
6118 lock = ldlm_handle2lock(lockh);
6119 LASSERT(lock != NULL);
6121 if (!ldlm_has_layout(lock))
6122 GOTO(out, rc = -EAGAIN);
6124 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
6125 PFID(&lli->lli_fid), inode);
6127 /* in case this is a caching lock and reinstate with new inode */
6128 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
6130 lock_res_and_lock(lock);
6131 lvb_ready = ldlm_is_lvb_ready(lock);
6132 unlock_res_and_lock(lock);
6134 /* checking lvb_ready is racy but this is okay. The worst case is
6135 * that multi processes may configure the file on the same time. */
6139 rc = ll_layout_fetch(inode, lock);
6143 /* for layout lock, lmm is stored in lock's lvb.
6144 * lvb_data is immutable if the lock is held so it's safe to access it
6147 * set layout to file. Unlikely this will fail as old layout was
6148 * surely eliminated */
6149 memset(&conf, 0, sizeof conf);
6150 conf.coc_opc = OBJECT_CONF_SET;
6151 conf.coc_inode = inode;
6152 conf.coc_lock = lock;
6153 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
6154 conf.u.coc_layout.lb_len = lock->l_lvb_len;
6155 rc = ll_layout_conf(inode, &conf);
6157 /* refresh layout failed, need to wait */
6158 wait_layout = rc == -EBUSY;
6161 LDLM_LOCK_PUT(lock);
6162 ldlm_lock_decref(lockh, mode);
6164 /* wait for IO to complete if it's still being used. */
6166 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
6167 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6169 memset(&conf, 0, sizeof conf);
6170 conf.coc_opc = OBJECT_CONF_WAIT;
6171 conf.coc_inode = inode;
6172 rc = ll_layout_conf(inode, &conf);
6176 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
6177 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
6183 * Issue layout intent RPC to MDS.
6184 * \param inode [in] file inode
6185 * \param intent [in] layout intent
6187 * \retval 0 on success
6188 * \retval < 0 error code
6190 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
6192 struct ll_inode_info *lli = ll_i2info(inode);
6193 struct ll_sb_info *sbi = ll_i2sbi(inode);
6194 struct md_op_data *op_data;
6195 struct lookup_intent it;
6196 struct ptlrpc_request *req;
6200 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
6201 0, 0, LUSTRE_OPC_ANY, NULL);
6202 if (IS_ERR(op_data))
6203 RETURN(PTR_ERR(op_data));
6205 op_data->op_data = intent;
6206 op_data->op_data_size = sizeof(*intent);
6208 memset(&it, 0, sizeof(it));
6209 it.it_op = IT_LAYOUT;
6210 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
6211 intent->li_opc == LAYOUT_INTENT_TRUNC)
6212 it.it_flags = FMODE_WRITE;
6214 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
6215 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6217 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
6218 &ll_md_blocking_ast, 0);
6219 if (it.it_request != NULL)
6220 ptlrpc_req_finished(it.it_request);
6221 it.it_request = NULL;
6223 ll_finish_md_op_data(op_data);
6225 /* set lock data in case this is a new lock */
6227 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
6229 ll_intent_drop_lock(&it);
6235 * This function checks if there exists a LAYOUT lock on the client side,
6236 * or enqueues it if it doesn't have one in cache.
6238 * This function will not hold layout lock so it may be revoked any time after
6239 * this function returns. Any operations depend on layout should be redone
6242 * This function should be called before lov_io_init() to get an uptodate
6243 * layout version, the caller should save the version number and after IO
6244 * is finished, this function should be called again to verify that layout
6245 * is not changed during IO time.
6247 int ll_layout_refresh(struct inode *inode, __u32 *gen)
6249 struct ll_inode_info *lli = ll_i2info(inode);
6250 struct ll_sb_info *sbi = ll_i2sbi(inode);
6251 struct lustre_handle lockh;
6252 struct layout_intent intent = {
6253 .li_opc = LAYOUT_INTENT_ACCESS,
6255 enum ldlm_mode mode;
6259 *gen = ll_layout_version_get(lli);
6260 if (!test_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags) ||
6261 *gen != CL_LAYOUT_GEN_NONE)
6265 LASSERT(fid_is_sane(ll_inode2fid(inode)));
6266 LASSERT(S_ISREG(inode->i_mode));
6268 /* take layout lock mutex to enqueue layout lock exclusively. */
6269 mutex_lock(&lli->lli_layout_mutex);
6272 /* mostly layout lock is caching on the local side, so try to
6273 * match it before grabbing layout lock mutex. */
6274 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
6275 LCK_CR | LCK_CW | LCK_PR |
6277 if (mode != 0) { /* hit cached lock */
6278 rc = ll_layout_lock_set(&lockh, mode, inode);
6284 rc = ll_layout_intent(inode, &intent);
6290 *gen = ll_layout_version_get(lli);
6291 mutex_unlock(&lli->lli_layout_mutex);
6297 * Issue layout intent RPC indicating where in a file an IO is about to write.
6299 * \param[in] inode file inode.
6300 * \param[in] ext write range with start offset of fille in bytes where
6301 * an IO is about to write, and exclusive end offset in
6304 * \retval 0 on success
6305 * \retval < 0 error code
6307 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
6308 struct lu_extent *ext)
6310 struct layout_intent intent = {
6312 .li_extent.e_start = ext->e_start,
6313 .li_extent.e_end = ext->e_end,
6318 rc = ll_layout_intent(inode, &intent);
6324 * This function send a restore request to the MDT
6326 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6328 struct hsm_user_request *hur;
6332 len = sizeof(struct hsm_user_request) +
6333 sizeof(struct hsm_user_item);
6334 OBD_ALLOC(hur, len);
6338 hur->hur_request.hr_action = HUA_RESTORE;
6339 hur->hur_request.hr_archive_id = 0;
6340 hur->hur_request.hr_flags = 0;
6341 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6342 sizeof(hur->hur_user_item[0].hui_fid));
6343 hur->hur_user_item[0].hui_extent.offset = offset;
6344 hur->hur_user_item[0].hui_extent.length = length;
6345 hur->hur_request.hr_itemcount = 1;
6346 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,