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>
47 #ifdef HAVE_LINUX_FILELOCK_HEADER
48 #include <linux/filelock.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <lustre_swab.h>
54 #include "cl_object.h"
55 #include "llite_internal.h"
56 #include "vvp_internal.h"
59 struct inode *sp_inode;
64 __u64 pa_data_version;
70 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
72 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
75 static struct ll_file_data *ll_file_data_get(void)
77 struct ll_file_data *fd;
79 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
83 fd->fd_write_failed = false;
84 pcc_file_init(&fd->fd_pcc_file);
89 static void ll_file_data_put(struct ll_file_data *fd)
92 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
96 * Packs all the attributes into @op_data for the CLOSE rpc.
98 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
99 struct obd_client_handle *och)
103 ll_prep_md_op_data(op_data, inode, NULL, NULL,
104 0, 0, LUSTRE_OPC_ANY, NULL);
106 op_data->op_attr.ia_mode = inode->i_mode;
107 op_data->op_attr.ia_atime = inode->i_atime;
108 op_data->op_attr.ia_mtime = inode->i_mtime;
109 op_data->op_attr.ia_ctime = inode->i_ctime;
110 /* In case of encrypted file without the key, visible size was rounded
111 * up to next LUSTRE_ENCRYPTION_UNIT_SIZE, and clear text size was
112 * stored into lli_lazysize in ll_merge_attr(), so set proper file size
113 * now that we are closing.
115 if (llcrypt_require_key(inode) == -ENOKEY &&
116 ll_i2info(inode)->lli_attr_valid & OBD_MD_FLLAZYSIZE)
117 op_data->op_attr.ia_size = ll_i2info(inode)->lli_lazysize;
119 op_data->op_attr.ia_size = i_size_read(inode);
120 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
121 ATTR_MTIME | ATTR_MTIME_SET |
123 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
124 op_data->op_attr_blocks = inode->i_blocks;
125 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
126 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
127 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
128 op_data->op_open_handle = och->och_open_handle;
130 if (och->och_flags & FMODE_WRITE &&
131 test_and_clear_bit(LLIF_DATA_MODIFIED,
132 &ll_i2info(inode)->lli_flags))
133 /* For HSM: if inode data has been modified, pack it so that
134 * MDT can set data dirty flag in the archive. */
135 op_data->op_bias |= MDS_DATA_MODIFIED;
141 * Perform a close, possibly with a bias.
142 * The meaning of "data" depends on the value of "bias".
144 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
145 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
148 static int ll_close_inode_openhandle(struct inode *inode,
149 struct obd_client_handle *och,
150 enum mds_op_bias bias, void *data)
152 struct obd_export *md_exp = ll_i2mdexp(inode);
153 const struct ll_inode_info *lli = ll_i2info(inode);
154 struct md_op_data *op_data;
155 struct ptlrpc_request *req = NULL;
159 if (class_exp2obd(md_exp) == NULL) {
160 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
161 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
165 OBD_ALLOC_PTR(op_data);
166 /* We leak openhandle and request here on error, but not much to be
167 * done in OOM case since app won't retry close on error either. */
169 GOTO(out, rc = -ENOMEM);
171 ll_prepare_close(inode, op_data, och);
173 case MDS_CLOSE_LAYOUT_MERGE:
174 /* merge blocks from the victim inode */
175 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
176 op_data->op_attr.ia_valid |= ATTR_SIZE;
177 op_data->op_xvalid |= OP_XVALID_BLOCKS;
179 case MDS_CLOSE_LAYOUT_SPLIT:
180 case MDS_CLOSE_LAYOUT_SWAP: {
181 struct split_param *sp = data;
183 LASSERT(data != NULL);
184 op_data->op_bias |= bias;
185 op_data->op_data_version = 0;
186 op_data->op_lease_handle = och->och_lease_handle;
187 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
188 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
189 op_data->op_mirror_id = sp->sp_mirror_id;
191 op_data->op_fid2 = *ll_inode2fid(data);
196 case MDS_CLOSE_RESYNC_DONE: {
197 struct ll_ioc_lease *ioc = data;
199 LASSERT(data != NULL);
200 op_data->op_attr_blocks +=
201 ioc->lil_count * op_data->op_attr_blocks;
202 op_data->op_attr.ia_valid |= ATTR_SIZE;
203 op_data->op_xvalid |= OP_XVALID_BLOCKS;
204 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
206 op_data->op_lease_handle = och->och_lease_handle;
207 op_data->op_data = &ioc->lil_ids[0];
208 op_data->op_data_size =
209 ioc->lil_count * sizeof(ioc->lil_ids[0]);
213 case MDS_PCC_ATTACH: {
214 struct pcc_param *param = data;
216 LASSERT(data != NULL);
217 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
218 op_data->op_archive_id = param->pa_archive_id;
219 op_data->op_data_version = param->pa_data_version;
220 op_data->op_lease_handle = och->och_lease_handle;
224 case MDS_HSM_RELEASE:
225 LASSERT(data != NULL);
226 op_data->op_bias |= MDS_HSM_RELEASE;
227 op_data->op_data_version = *(__u64 *)data;
228 op_data->op_lease_handle = och->och_lease_handle;
229 op_data->op_attr.ia_valid |= ATTR_SIZE;
230 op_data->op_xvalid |= OP_XVALID_BLOCKS;
234 LASSERT(data == NULL);
238 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
239 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
240 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
241 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
243 rc = md_close(md_exp, op_data, och->och_mod, &req);
244 if (rc != 0 && rc != -EINTR)
245 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
246 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
248 if (rc == 0 && op_data->op_bias & bias) {
249 struct mdt_body *body;
251 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
252 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
255 if (bias & MDS_PCC_ATTACH) {
256 struct pcc_param *param = data;
258 param->pa_layout_gen = body->mbo_layout_gen;
262 ll_finish_md_op_data(op_data);
266 md_clear_open_replay_data(md_exp, och);
267 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
270 ptlrpc_req_finished(req); /* This is close request */
274 int ll_md_real_close(struct inode *inode, fmode_t fmode)
276 struct ll_inode_info *lli = ll_i2info(inode);
277 struct obd_client_handle **och_p;
278 struct obd_client_handle *och;
283 if (fmode & FMODE_WRITE) {
284 och_p = &lli->lli_mds_write_och;
285 och_usecount = &lli->lli_open_fd_write_count;
286 } else if (fmode & FMODE_EXEC) {
287 och_p = &lli->lli_mds_exec_och;
288 och_usecount = &lli->lli_open_fd_exec_count;
290 LASSERT(fmode & FMODE_READ);
291 och_p = &lli->lli_mds_read_och;
292 och_usecount = &lli->lli_open_fd_read_count;
295 mutex_lock(&lli->lli_och_mutex);
296 if (*och_usecount > 0) {
297 /* There are still users of this handle, so skip
299 mutex_unlock(&lli->lli_och_mutex);
305 mutex_unlock(&lli->lli_och_mutex);
308 /* There might be a race and this handle may already
310 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
316 static int ll_md_close(struct inode *inode, struct file *file)
318 union ldlm_policy_data policy = {
319 .l_inodebits = { MDS_INODELOCK_OPEN },
321 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
322 struct ll_file_data *fd = file->private_data;
323 struct ll_inode_info *lli = ll_i2info(inode);
324 struct lustre_handle lockh;
325 enum ldlm_mode lockmode;
329 /* clear group lock, if present */
330 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
331 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
333 mutex_lock(&lli->lli_och_mutex);
334 if (fd->fd_lease_och != NULL) {
336 struct obd_client_handle *lease_och;
338 lease_och = fd->fd_lease_och;
339 fd->fd_lease_och = NULL;
340 mutex_unlock(&lli->lli_och_mutex);
342 /* Usually the lease is not released when the
343 * application crashed, we need to release here. */
344 rc = ll_lease_close(lease_och, inode, &lease_broken);
346 mutex_lock(&lli->lli_och_mutex);
348 CDEBUG_LIMIT(rc ? D_ERROR : D_INODE,
349 "Clean up lease "DFID" %d/%d\n",
350 PFID(&lli->lli_fid), rc, lease_broken);
353 if (fd->fd_och != NULL) {
354 struct obd_client_handle *och;
358 mutex_unlock(&lli->lli_och_mutex);
360 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
364 /* Let's see if we have good enough OPEN lock on the file and if
365 we can skip talking to MDS */
366 if (fd->fd_omode & FMODE_WRITE) {
368 LASSERT(lli->lli_open_fd_write_count);
369 lli->lli_open_fd_write_count--;
370 } else if (fd->fd_omode & FMODE_EXEC) {
372 LASSERT(lli->lli_open_fd_exec_count);
373 lli->lli_open_fd_exec_count--;
376 LASSERT(lli->lli_open_fd_read_count);
377 lli->lli_open_fd_read_count--;
379 mutex_unlock(&lli->lli_och_mutex);
381 /* LU-4398: do not cache write open lock if the file has exec bit */
382 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
383 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
384 LDLM_IBITS, &policy, lockmode, &lockh))
385 rc = ll_md_real_close(inode, fd->fd_omode);
388 file->private_data = NULL;
389 ll_file_data_put(fd);
394 /* While this returns an error code, fput() the caller does not, so we need
395 * to make every effort to clean up all of our state here. Also, applications
396 * rarely check close errors and even if an error is returned they will not
397 * re-try the close call.
399 int ll_file_release(struct inode *inode, struct file *file)
401 struct ll_file_data *fd;
402 struct ll_sb_info *sbi = ll_i2sbi(inode);
403 struct ll_inode_info *lli = ll_i2info(inode);
404 ktime_t kstart = ktime_get();
409 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
410 PFID(ll_inode2fid(inode)), inode);
412 fd = file->private_data;
415 /* The last ref on @file, maybe not the the owner pid of statahead,
416 * because parent and child process can share the same file handle. */
417 if (S_ISDIR(inode->i_mode) &&
418 (lli->lli_opendir_key == fd || fd->fd_sai))
419 ll_deauthorize_statahead(inode, fd);
421 if (is_root_inode(inode)) {
422 file->private_data = NULL;
423 ll_file_data_put(fd);
427 pcc_file_release(inode, file);
429 if (!S_ISDIR(inode->i_mode)) {
430 if (lli->lli_clob != NULL)
431 lov_read_and_clear_async_rc(lli->lli_clob);
432 lli->lli_async_rc = 0;
435 lli->lli_close_fd_time = ktime_get();
437 rc = ll_md_close(inode, file);
439 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
440 libcfs_debug_dumplog();
443 if (!rc && !is_root_inode(inode))
444 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
445 ktime_us_delta(ktime_get(), kstart));
449 static inline int ll_dom_readpage(void *data, struct page *page)
451 /* since ll_dom_readpage is a page cache helper, it is safe to assume
452 * mapping and host pointers are set here
455 struct niobuf_local *lnb = data;
459 inode = page2inode(page);
461 kaddr = kmap_atomic(page);
462 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
463 if (lnb->lnb_len < PAGE_SIZE)
464 memset(kaddr + lnb->lnb_len, 0,
465 PAGE_SIZE - lnb->lnb_len);
466 kunmap_atomic(kaddr);
468 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
469 if (!llcrypt_has_encryption_key(inode)) {
470 CDEBUG(D_SEC, "no enc key for "DFID"\n",
471 PFID(ll_inode2fid(inode)));
474 unsigned int offs = 0;
476 while (offs < PAGE_SIZE) {
477 /* decrypt only if page is not empty */
478 if (memcmp(page_address(page) + offs,
479 page_address(ZERO_PAGE(0)),
480 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
483 rc = llcrypt_decrypt_pagecache_blocks(page,
484 LUSTRE_ENCRYPTION_UNIT_SIZE,
489 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
494 flush_dcache_page(page);
495 SetPageUptodate(page);
502 #ifdef HAVE_READ_CACHE_PAGE_WANTS_FILE
503 static inline int ll_dom_read_folio(struct file *file, struct folio *folio0)
505 return ll_dom_readpage(file->private_data, folio_page(folio0, 0));
508 #define ll_dom_read_folio ll_dom_readpage
511 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
515 struct ll_inode_info *lli = ll_i2info(inode);
516 struct cl_object *obj = lli->lli_clob;
517 struct address_space *mapping = inode->i_mapping;
519 struct niobuf_remote *rnb;
520 struct mdt_body *body;
522 unsigned long index, start;
523 struct niobuf_local lnb;
532 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
536 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
537 if (rnb == NULL || rnb->rnb_len == 0)
540 /* LU-11595: Server may return whole file and that is OK always or
541 * it may return just file tail and its offset must be aligned with
542 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
543 * smaller then offset may be not aligned and that data is just ignored.
545 if (rnb->rnb_offset & ~PAGE_MASK)
548 /* Server returns whole file or just file tail if it fills in reply
549 * buffer, in both cases total size should be equal to the file size.
551 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
552 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
553 !(inode && IS_ENCRYPTED(inode))) {
554 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
555 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
556 rnb->rnb_len, body->mbo_dom_size);
560 env = cl_env_get(&refcheck);
563 io = vvp_env_thread_io(env);
565 io->ci_ignore_layout = 1;
566 rc = cl_io_init(env, io, CIT_MISC, obj);
570 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
571 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
573 data = (char *)rnb + sizeof(*rnb);
575 lnb.lnb_file_offset = rnb->rnb_offset;
576 start = lnb.lnb_file_offset >> PAGE_SHIFT;
578 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
579 lnb.lnb_page_offset = 0;
581 struct cl_page *page;
583 lnb.lnb_data = data + (index << PAGE_SHIFT);
584 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
585 if (lnb.lnb_len > PAGE_SIZE)
586 lnb.lnb_len = PAGE_SIZE;
588 vmpage = ll_read_cache_page(mapping, index + start,
589 ll_dom_read_folio, &lnb);
590 if (IS_ERR(vmpage)) {
591 CWARN("%s: cannot fill page %lu for "DFID
592 " with data: rc = %li\n",
593 ll_i2sbi(inode)->ll_fsname, index + start,
594 PFID(lu_object_fid(&obj->co_lu)),
599 if (vmpage->mapping == NULL) {
602 /* page was truncated */
605 /* attach VM page to CL page cache */
606 page = cl_page_find(env, obj, vmpage->index, vmpage,
609 ClearPageUptodate(vmpage);
614 SetPageUptodate(vmpage);
615 cl_page_put(env, page);
619 } while (rnb->rnb_len > (index << PAGE_SHIFT));
623 cl_env_put(env, &refcheck);
628 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
629 struct lookup_intent *itp)
631 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
632 struct dentry *parent = de->d_parent;
635 struct md_op_data *op_data;
636 struct ptlrpc_request *req = NULL;
640 LASSERT(parent != NULL);
641 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
643 /* if server supports open-by-fid, or file name is invalid, don't pack
644 * name in open request */
645 if (CFS_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
646 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
648 len = de->d_name.len;
649 name = kmalloc(len + 1, GFP_NOFS);
654 spin_lock(&de->d_lock);
655 if (len != de->d_name.len) {
656 spin_unlock(&de->d_lock);
660 memcpy(name, de->d_name.name, len);
662 spin_unlock(&de->d_lock);
664 if (!lu_name_is_valid_2(name, len)) {
670 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
671 name, len, 0, LUSTRE_OPC_OPEN, NULL);
672 if (IS_ERR(op_data)) {
674 RETURN(PTR_ERR(op_data));
676 op_data->op_data = lmm;
677 op_data->op_data_size = lmmsize;
679 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_OPEN_DELAY, cfs_fail_val);
681 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
682 &ll_md_blocking_ast, 0);
684 ll_finish_md_op_data(op_data);
686 /* reason for keep own exit path - don`t flood log
687 * with messages with -ESTALE errors.
689 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
690 it_open_error(DISP_OPEN_OPEN, itp))
692 ll_release_openhandle(de, itp);
696 if (it_disposition(itp, DISP_LOOKUP_NEG))
697 GOTO(out, rc = -ENOENT);
699 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
700 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
701 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
705 rc = ll_prep_inode(&de->d_inode, &req->rq_pill, NULL, itp);
707 if (!rc && itp->it_lock_mode) {
710 /* If we got a lock back and it has a LOOKUP bit set,
711 * make sure the dentry is marked as valid so we can find it.
712 * We don't need to care about actual hashing since other bits
713 * of kernel will deal with that later.
715 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
716 if (bits & MDS_INODELOCK_LOOKUP)
717 d_lustre_revalidate(de);
719 /* if DoM bit returned along with LAYOUT bit then there
720 * can be read-on-open data returned.
722 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
723 ll_dom_finish_open(de->d_inode, req);
725 /* open may not fetch LOOKUP lock, update dir depth and default LMV
728 if (!rc && S_ISDIR(de->d_inode->i_mode))
729 ll_update_dir_depth_dmv(parent->d_inode, de);
732 ptlrpc_req_finished(req);
733 ll_intent_drop_lock(itp);
735 /* We did open by fid, but by the time we got to the server, the object
736 * disappeared. This is possible if the object was unlinked, but it's
737 * also possible if the object was unlinked by a rename. In the case
738 * of an object renamed over our existing one, we can't fail this open.
739 * O_CREAT also goes through this path if we had an existing dentry,
740 * and it's obviously wrong to return ENOENT for O_CREAT.
742 * Instead let's return -ESTALE, and the VFS will retry the open with
743 * LOOKUP_REVAL, which we catch in ll_revalidate_dentry and fail to
744 * revalidate, causing a lookup. This causes extra lookups in the case
745 * where we had a dentry in cache but the file is being unlinked and we
746 * lose the race with unlink, but this should be very rare.
754 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
755 struct obd_client_handle *och)
757 struct mdt_body *body;
759 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
760 och->och_open_handle = body->mbo_open_handle;
761 och->och_fid = body->mbo_fid1;
762 och->och_lease_handle.cookie = it->it_lock_handle;
763 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
764 och->och_flags = it->it_flags;
766 return md_set_open_replay_data(md_exp, och, it);
769 static int ll_local_open(struct file *file, struct lookup_intent *it,
770 struct ll_file_data *fd, struct obd_client_handle *och)
772 struct inode *inode = file_inode(file);
775 LASSERT(!file->private_data);
782 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
787 file->private_data = fd;
788 ll_readahead_init(inode, &fd->fd_ras);
789 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
794 void ll_track_file_opens(struct inode *inode)
796 struct ll_inode_info *lli = ll_i2info(inode);
797 struct ll_sb_info *sbi = ll_i2sbi(inode);
799 /* do not skew results with delays from never-opened inodes */
800 if (ktime_to_ns(lli->lli_close_fd_time))
801 ll_stats_ops_tally(sbi, LPROC_LL_INODE_OPCLTM,
802 ktime_us_delta(ktime_get(), lli->lli_close_fd_time));
804 if (ktime_after(ktime_get(),
805 ktime_add_ms(lli->lli_close_fd_time,
806 sbi->ll_oc_max_ms))) {
807 lli->lli_open_fd_count = 1;
808 lli->lli_close_fd_time = ns_to_ktime(0);
810 lli->lli_open_fd_count++;
813 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_OCOUNT,
814 lli->lli_open_fd_count);
817 /* Open a file, and (for the very first open) create objects on the OSTs at
818 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
819 * creation or open until ll_lov_setstripe() ioctl is called.
821 * If we already have the stripe MD locally then we don't request it in
822 * md_open(), by passing a lmm_size = 0.
824 * It is up to the application to ensure no other processes open this file
825 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
826 * used. We might be able to avoid races of that sort by getting lli_open_sem
827 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
828 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
830 int ll_file_open(struct inode *inode, struct file *file)
832 struct ll_inode_info *lli = ll_i2info(inode);
833 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
834 .it_flags = file->f_flags };
835 struct obd_client_handle **och_p = NULL;
836 __u64 *och_usecount = NULL;
837 struct ll_file_data *fd;
838 ktime_t kstart = ktime_get();
842 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
843 PFID(ll_inode2fid(inode)), inode, file->f_flags);
845 it = file->private_data; /* XXX: compat macro */
846 file->private_data = NULL; /* prevent ll_local_open assertion */
848 if (S_ISREG(inode->i_mode)) {
849 rc = ll_file_open_encrypt(inode, file);
851 if (it && it->it_disposition)
852 ll_release_openhandle(file_dentry(file), it);
853 GOTO(out_nofiledata, rc);
857 fd = ll_file_data_get();
859 GOTO(out_nofiledata, rc = -ENOMEM);
862 if (S_ISDIR(inode->i_mode))
863 ll_authorize_statahead(inode, fd);
865 ll_track_file_opens(inode);
866 if (is_root_inode(inode)) {
867 file->private_data = fd;
871 if (!it || !it->it_disposition) {
872 /* Convert f_flags into access mode. We cannot use file->f_mode,
873 * because everything but O_ACCMODE mask was stripped from
875 if ((oit.it_flags + 1) & O_ACCMODE)
877 if (file->f_flags & O_TRUNC)
878 oit.it_flags |= FMODE_WRITE;
880 /* kernel only call f_op->open in dentry_open. filp_open calls
881 * dentry_open after call to open_namei that checks permissions.
882 * Only nfsd_open call dentry_open directly without checking
883 * permissions and because of that this code below is safe.
885 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
886 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
888 /* We do not want O_EXCL here, presumably we opened the file
889 * already? XXX - NFS implications? */
890 oit.it_flags &= ~O_EXCL;
892 /* bug20584, if "it_flags" contains O_CREAT, the file will be
893 * created if necessary, then "IT_CREAT" should be set to keep
894 * consistent with it */
895 if (oit.it_flags & O_CREAT)
896 oit.it_op |= IT_CREAT;
902 /* Let's see if we have file open on MDS already. */
903 if (it->it_flags & FMODE_WRITE) {
904 och_p = &lli->lli_mds_write_och;
905 och_usecount = &lli->lli_open_fd_write_count;
906 } else if (it->it_flags & FMODE_EXEC) {
907 och_p = &lli->lli_mds_exec_och;
908 och_usecount = &lli->lli_open_fd_exec_count;
910 och_p = &lli->lli_mds_read_och;
911 och_usecount = &lli->lli_open_fd_read_count;
914 mutex_lock(&lli->lli_och_mutex);
915 if (*och_p) { /* Open handle is present */
916 if (it_disposition(it, DISP_OPEN_OPEN)) {
917 /* Well, there's extra open request that we do not need,
918 * let's close it somehow. This will decref request. */
919 rc = it_open_error(DISP_OPEN_OPEN, it);
921 mutex_unlock(&lli->lli_och_mutex);
922 GOTO(out_openerr, rc);
925 ll_release_openhandle(file_dentry(file), it);
929 rc = ll_local_open(file, it, fd, NULL);
932 mutex_unlock(&lli->lli_och_mutex);
933 GOTO(out_openerr, rc);
936 LASSERT(*och_usecount == 0);
937 if (!it->it_disposition) {
938 struct dentry *dentry = file_dentry(file);
939 struct ll_sb_info *sbi = ll_i2sbi(inode);
940 int open_threshold = sbi->ll_oc_thrsh_count;
942 /* We cannot just request lock handle now, new ELC code
943 * means that one of other OPEN locks for this file
944 * could be cancelled, and since blocking ast handler
945 * would attempt to grab och_mutex as well, that would
946 * result in a deadlock
948 mutex_unlock(&lli->lli_och_mutex);
950 * Normally called under two situations:
951 * 1. fhandle / NFS export.
952 * 2. A race/condition on MDS resulting in no open
953 * handle to be returned from LOOKUP|OPEN request,
954 * for example if the target entry was a symlink.
956 * For NFSv3 we need to always cache the open lock
957 * for pre 5.5 Linux kernels.
959 * After reaching number of opens of this inode
960 * we always ask for an open lock on it to handle
961 * bad userspace actors that open and close files
962 * in a loop for absolutely no good reason
964 /* fhandle / NFS path. */
965 if (lli->lli_open_thrsh_count != UINT_MAX)
966 open_threshold = lli->lli_open_thrsh_count;
968 if (filename_is_volatile(dentry->d_name.name,
971 /* There really is nothing here, but this
972 * make this more readable I think.
973 * We do not want openlock for volatile
974 * files under any circumstances
976 } else if (open_threshold > 0) {
977 /* Take MDS_OPEN_LOCK with many opens */
978 if (lli->lli_open_fd_count >= open_threshold)
979 it->it_flags |= MDS_OPEN_LOCK;
981 /* If this is open after we just closed */
982 else if (ktime_before(ktime_get(),
983 ktime_add_ms(lli->lli_close_fd_time,
984 sbi->ll_oc_thrsh_ms)))
985 it->it_flags |= MDS_OPEN_LOCK;
989 * Always specify MDS_OPEN_BY_FID because we don't want
990 * to get file with different fid.
992 it->it_flags |= MDS_OPEN_BY_FID;
993 rc = ll_intent_file_open(dentry, NULL, 0, it);
995 GOTO(out_openerr, rc);
999 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
1001 GOTO(out_och_free, rc = -ENOMEM);
1005 /* md_intent_lock() didn't get a request ref if there was an
1006 * open error, so don't do cleanup on the request here
1008 /* XXX (green): Should not we bail out on any error here, not
1009 * just open error? */
1010 rc = it_open_error(DISP_OPEN_OPEN, it);
1012 GOTO(out_och_free, rc);
1014 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
1015 "inode %px: disposition %x, status %d\n", inode,
1016 it_disposition(it, ~0), it->it_status);
1018 rc = ll_local_open(file, it, fd, *och_p);
1020 GOTO(out_och_free, rc);
1023 rc = pcc_file_open(inode, file);
1025 GOTO(out_och_free, rc);
1027 mutex_unlock(&lli->lli_och_mutex);
1031 /* Must do this outside lli_och_mutex lock to prevent deadlock where
1032 different kind of OPEN lock for this same inode gets cancelled
1033 by ldlm_cancel_lru */
1034 if (!S_ISREG(inode->i_mode))
1035 GOTO(out_och_free, rc);
1036 cl_lov_delay_create_clear(&file->f_flags);
1037 GOTO(out_och_free, rc);
1041 if (och_p && *och_p) {
1042 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
1043 *och_p = NULL; /* OBD_FREE writes some magic there */
1046 mutex_unlock(&lli->lli_och_mutex);
1049 if (lli->lli_opendir_key == fd)
1050 ll_deauthorize_statahead(inode, fd);
1053 ll_file_data_put(fd);
1055 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
1056 ktime_us_delta(ktime_get(), kstart));
1060 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
1061 ptlrpc_req_finished(it->it_request);
1062 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1068 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
1069 struct ldlm_lock_desc *desc, void *data, int flag)
1072 struct lustre_handle lockh;
1076 case LDLM_CB_BLOCKING:
1077 ldlm_lock2handle(lock, &lockh);
1078 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
1080 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1084 case LDLM_CB_CANCELING:
1092 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1093 * and save it as fd->fd_och so as to force client to reopen the file even
1094 * if it has an open lock in cache already.
1096 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1097 struct lustre_handle *old_open_handle)
1099 struct ll_inode_info *lli = ll_i2info(inode);
1100 struct ll_file_data *fd = file->private_data;
1101 struct obd_client_handle **och_p;
1102 __u64 *och_usecount;
1106 /* Get the openhandle of the file */
1107 mutex_lock(&lli->lli_och_mutex);
1108 if (fd->fd_lease_och != NULL)
1109 GOTO(out_unlock, rc = -EBUSY);
1111 if (fd->fd_och == NULL) {
1112 if (file->f_mode & FMODE_WRITE) {
1113 LASSERT(lli->lli_mds_write_och != NULL);
1114 och_p = &lli->lli_mds_write_och;
1115 och_usecount = &lli->lli_open_fd_write_count;
1117 LASSERT(lli->lli_mds_read_och != NULL);
1118 och_p = &lli->lli_mds_read_och;
1119 och_usecount = &lli->lli_open_fd_read_count;
1122 if (*och_usecount > 1)
1123 GOTO(out_unlock, rc = -EBUSY);
1125 fd->fd_och = *och_p;
1130 *old_open_handle = fd->fd_och->och_open_handle;
1134 mutex_unlock(&lli->lli_och_mutex);
1139 * Release ownership on lli_mds_*_och when putting back a file lease.
1141 static int ll_lease_och_release(struct inode *inode, struct file *file)
1143 struct ll_inode_info *lli = ll_i2info(inode);
1144 struct ll_file_data *fd = file->private_data;
1145 struct obd_client_handle **och_p;
1146 struct obd_client_handle *old_och = NULL;
1147 __u64 *och_usecount;
1151 mutex_lock(&lli->lli_och_mutex);
1152 if (file->f_mode & FMODE_WRITE) {
1153 och_p = &lli->lli_mds_write_och;
1154 och_usecount = &lli->lli_open_fd_write_count;
1156 och_p = &lli->lli_mds_read_och;
1157 och_usecount = &lli->lli_open_fd_read_count;
1160 /* The file may have been open by another process (broken lease) so
1161 * *och_p is not NULL. In this case we should simply increase usecount
1164 if (*och_p != NULL) {
1165 old_och = fd->fd_och;
1168 *och_p = fd->fd_och;
1172 mutex_unlock(&lli->lli_och_mutex);
1174 if (old_och != NULL)
1175 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1181 * Acquire a lease and open the file.
1183 static struct obd_client_handle *
1184 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1187 struct lookup_intent it = { .it_op = IT_OPEN };
1188 struct ll_sb_info *sbi = ll_i2sbi(inode);
1189 struct md_op_data *op_data;
1190 struct ptlrpc_request *req = NULL;
1191 struct lustre_handle old_open_handle = { 0 };
1192 struct obd_client_handle *och = NULL;
1197 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1198 RETURN(ERR_PTR(-EINVAL));
1201 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1202 RETURN(ERR_PTR(-EPERM));
1204 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1206 RETURN(ERR_PTR(rc));
1211 RETURN(ERR_PTR(-ENOMEM));
1213 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1214 LUSTRE_OPC_ANY, NULL);
1215 if (IS_ERR(op_data))
1216 GOTO(out, rc = PTR_ERR(op_data));
1218 /* To tell the MDT this openhandle is from the same owner */
1219 op_data->op_open_handle = old_open_handle;
1221 it.it_flags = fmode | open_flags;
1222 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1223 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1224 &ll_md_blocking_lease_ast,
1225 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1226 * it can be cancelled which may mislead applications that the lease is
1228 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1229 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1230 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1231 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1232 ll_finish_md_op_data(op_data);
1233 ptlrpc_req_finished(req);
1235 GOTO(out_release_it, rc);
1237 if (it_disposition(&it, DISP_LOOKUP_NEG))
1238 GOTO(out_release_it, rc = -ENOENT);
1240 rc = it_open_error(DISP_OPEN_OPEN, &it);
1242 GOTO(out_release_it, rc);
1244 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1245 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1247 GOTO(out_release_it, rc);
1249 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1250 GOTO(out_close, rc = -EOPNOTSUPP);
1252 /* already get lease, handle lease lock */
1253 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1254 if (!it.it_lock_mode ||
1255 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1256 /* open lock must return for lease */
1257 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1258 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1260 GOTO(out_close, rc = -EPROTO);
1263 ll_intent_release(&it);
1267 /* Cancel open lock */
1268 if (it.it_lock_mode != 0) {
1269 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1271 it.it_lock_mode = 0;
1272 och->och_lease_handle.cookie = 0ULL;
1274 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1276 CERROR("%s: error closing file "DFID": %d\n",
1277 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1278 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1280 ll_intent_release(&it);
1284 RETURN(ERR_PTR(rc));
1288 * Check whether a layout swap can be done between two inodes.
1290 * \param[in] inode1 First inode to check
1291 * \param[in] inode2 Second inode to check
1293 * \retval 0 on success, layout swap can be performed between both inodes
1294 * \retval negative error code if requirements are not met
1296 static int ll_check_swap_layouts_validity(struct inode *inode1,
1297 struct inode *inode2)
1299 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1302 if (inode_permission(&nop_mnt_idmap, inode1, MAY_WRITE) ||
1303 inode_permission(&nop_mnt_idmap, inode2, MAY_WRITE))
1306 if (inode1->i_sb != inode2->i_sb)
1312 static int ll_swap_layouts_close(struct obd_client_handle *och,
1313 struct inode *inode, struct inode *inode2)
1315 const struct lu_fid *fid1 = ll_inode2fid(inode);
1316 const struct lu_fid *fid2;
1320 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1321 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1323 rc = ll_check_swap_layouts_validity(inode, inode2);
1325 GOTO(out_free_och, rc);
1327 /* We now know that inode2 is a lustre inode */
1328 fid2 = ll_inode2fid(inode2);
1330 rc = lu_fid_cmp(fid1, fid2);
1332 GOTO(out_free_och, rc = -EINVAL);
1334 /* Close the file and {swap,merge} layouts between inode & inode2.
1335 * NB: local lease handle is released in mdc_close_intent_pack()
1336 * because we still need it to pack l_remote_handle to MDT. */
1337 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1340 och = NULL; /* freed in ll_close_inode_openhandle() */
1350 * Release lease and close the file.
1351 * It will check if the lease has ever broken.
1353 static int ll_lease_close_intent(struct obd_client_handle *och,
1354 struct inode *inode,
1355 bool *lease_broken, enum mds_op_bias bias,
1358 struct ldlm_lock *lock;
1359 bool cancelled = true;
1363 lock = ldlm_handle2lock(&och->och_lease_handle);
1365 lock_res_and_lock(lock);
1366 cancelled = ldlm_is_cancel(lock);
1367 unlock_res_and_lock(lock);
1368 LDLM_LOCK_PUT(lock);
1371 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1372 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1374 if (lease_broken != NULL)
1375 *lease_broken = cancelled;
1377 if (!cancelled && !bias)
1378 ldlm_cli_cancel(&och->och_lease_handle, 0);
1380 if (cancelled) { /* no need to excute intent */
1385 rc = ll_close_inode_openhandle(inode, och, bias, data);
1389 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1392 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1396 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1398 static int ll_lease_file_resync(struct obd_client_handle *och,
1399 struct inode *inode, void __user *uarg)
1401 struct ll_sb_info *sbi = ll_i2sbi(inode);
1402 struct md_op_data *op_data;
1403 struct ll_ioc_lease_id ioc;
1404 __u64 data_version_unused;
1408 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1409 LUSTRE_OPC_ANY, NULL);
1410 if (IS_ERR(op_data))
1411 RETURN(PTR_ERR(op_data));
1413 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
1416 /* before starting file resync, it's necessary to clean up page cache
1417 * in client memory, otherwise once the layout version is increased,
1418 * writing back cached data will be denied the OSTs. */
1419 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1423 op_data->op_lease_handle = och->och_lease_handle;
1424 op_data->op_mirror_id = ioc.lil_mirror_id;
1425 rc = md_file_resync(sbi->ll_md_exp, op_data);
1431 ll_finish_md_op_data(op_data);
1435 static int ll_merge_attr_nolock(const struct lu_env *env, struct inode *inode)
1437 struct ll_inode_info *lli = ll_i2info(inode);
1438 struct cl_object *obj = lli->lli_clob;
1439 struct cl_attr *attr = vvp_env_thread_attr(env);
1447 /* Merge timestamps the most recently obtained from MDS with
1448 * timestamps obtained from OSTs.
1450 * Do not overwrite atime of inode because it may be refreshed
1451 * by file_accessed() function. If the read was served by cache
1452 * data, there is no RPC to be sent so that atime may not be
1453 * transferred to OSTs at all. MDT only updates atime at close time
1454 * if it's at least 'mdd.*.atime_diff' older.
1455 * All in all, the atime in Lustre does not strictly comply with
1456 * POSIX. Solving this problem needs to send an RPC to MDT for each
1457 * read, this will hurt performance.
1459 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1460 inode->i_atime.tv_sec < lli->lli_atime)
1461 inode->i_atime.tv_sec = lli->lli_atime;
1463 inode->i_mtime.tv_sec = lli->lli_mtime;
1464 inode->i_ctime.tv_sec = lli->lli_ctime;
1466 mtime = inode->i_mtime.tv_sec;
1467 atime = inode->i_atime.tv_sec;
1468 ctime = inode->i_ctime.tv_sec;
1470 cl_object_attr_lock(obj);
1471 if (CFS_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1474 rc = cl_object_attr_get(env, obj, attr);
1475 cl_object_attr_unlock(obj);
1478 GOTO(out, rc = (rc == -ENODATA ? 0 : rc));
1480 if (atime < attr->cat_atime)
1481 atime = attr->cat_atime;
1483 if (ctime < attr->cat_ctime)
1484 ctime = attr->cat_ctime;
1486 if (mtime < attr->cat_mtime)
1487 mtime = attr->cat_mtime;
1489 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu i_blocks %llu\n",
1490 PFID(&lli->lli_fid), attr->cat_size, attr->cat_blocks);
1492 if (llcrypt_require_key(inode) == -ENOKEY) {
1493 /* Without the key, round up encrypted file size to next
1494 * LUSTRE_ENCRYPTION_UNIT_SIZE. Clear text size is put in
1495 * lli_lazysize for proper file size setting at close time.
1497 lli->lli_attr_valid |= OBD_MD_FLLAZYSIZE;
1498 lli->lli_lazysize = attr->cat_size;
1499 attr->cat_size = round_up(attr->cat_size,
1500 LUSTRE_ENCRYPTION_UNIT_SIZE);
1502 i_size_write(inode, attr->cat_size);
1503 inode->i_blocks = attr->cat_blocks;
1505 inode->i_mtime.tv_sec = mtime;
1506 inode->i_atime.tv_sec = atime;
1507 inode->i_ctime.tv_sec = ctime;
1514 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1518 ll_inode_size_lock(inode);
1519 rc = ll_merge_attr_nolock(env, inode);
1520 ll_inode_size_unlock(inode);
1525 /* Use to update size and blocks on inode for LSOM if there is no contention */
1526 int ll_merge_attr_try(const struct lu_env *env, struct inode *inode)
1530 if (ll_inode_size_trylock(inode)) {
1531 rc = ll_merge_attr_nolock(env, inode);
1532 ll_inode_size_unlock(inode);
1539 * Set designated mirror for I/O.
1541 * So far only read, write, and truncated can support to issue I/O to
1542 * designated mirror.
1544 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1546 struct ll_file_data *fd = file->private_data;
1548 /* clear layout version for generic(non-resync) I/O in case it carries
1549 * stale layout version due to I/O restart */
1550 io->ci_layout_version = 0;
1552 /* FLR: disable non-delay for designated mirror I/O because obviously
1553 * only one mirror is available */
1554 if (fd->fd_designated_mirror > 0) {
1556 io->ci_designated_mirror = fd->fd_designated_mirror;
1557 io->ci_layout_version = fd->fd_layout_version;
1560 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1561 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1565 * This is relatime_need_update() from Linux 5.17, which is not exported.
1567 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1568 struct timespec64 now)
1571 if (!(mnt->mnt_flags & MNT_RELATIME))
1574 * Is mtime younger than atime? If yes, update atime:
1576 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1579 * Is ctime younger than atime? If yes, update atime:
1581 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1585 * Is the previous atime value older than a day? If yes,
1588 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1591 * Good, we can skip the atime update:
1597 * Very similar to kernel function: !__atime_needs_update()
1599 static bool file_is_noatime(const struct file *file)
1601 struct vfsmount *mnt = file->f_path.mnt;
1602 struct inode *inode = file_inode((struct file *)file);
1603 struct timespec64 now;
1605 if (file->f_flags & O_NOATIME)
1608 if (inode->i_flags & S_NOATIME)
1611 if (IS_NOATIME(inode))
1614 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1617 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1620 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1623 now = current_time(inode);
1625 if (!relatime_need_update(mnt, inode, now))
1631 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1632 struct vvp_io_args *args)
1634 struct inode *inode = file_inode(file);
1635 struct ll_file_data *fd = file->private_data;
1636 int flags = vvp_io_args_flags(file, args);
1638 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1639 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1641 if (iot == CIT_WRITE) {
1642 io->u.ci_wr.wr_append = iocb_ki_flags_check(flags, APPEND);
1643 io->u.ci_wr.wr_sync = !!(iocb_ki_flags_check(flags, SYNC) ||
1644 iocb_ki_flags_check(flags, APPEND) ||
1645 iocb_ki_flags_check(flags, DSYNC) ||
1650 io->ci_iocb_nowait = iocb_ki_flags_check(flags, NOWAIT);
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 (iocb_ki_flags_check(flags, 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);
1667 /* unaligned DIO has compat issues with some older servers, but we find
1668 * out if there are such servers while setting up the IO, so it starts
1671 io->ci_allow_unaligned_dio = true;
1673 ll_io_set_mirror(io, file);
1676 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1679 struct ll_inode_info *lli = ll_i2info(inode);
1680 struct ll_sb_info *sbi = ll_i2sbi(inode);
1681 enum obd_heat_type sample_type;
1682 enum obd_heat_type iobyte_type;
1683 __u64 now = ktime_get_real_seconds();
1685 if (!ll_sbi_has_file_heat(sbi) ||
1686 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1689 if (iot == CIT_READ) {
1690 sample_type = OBD_HEAT_READSAMPLE;
1691 iobyte_type = OBD_HEAT_READBYTE;
1692 } else if (iot == CIT_WRITE) {
1693 sample_type = OBD_HEAT_WRITESAMPLE;
1694 iobyte_type = OBD_HEAT_WRITEBYTE;
1699 spin_lock(&lli->lli_heat_lock);
1700 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1701 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1702 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1703 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1704 spin_unlock(&lli->lli_heat_lock);
1708 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1709 struct file *file, enum cl_io_type iot,
1710 loff_t *ppos, size_t bytes)
1712 struct vvp_io *vio = vvp_env_io(env);
1713 struct inode *inode = file_inode(file);
1714 struct ll_inode_info *lli = ll_i2info(inode);
1715 struct ll_sb_info *sbi = ll_i2sbi(inode);
1716 struct ll_file_data *fd = file->private_data;
1717 struct range_lock range;
1719 bool range_locked = false;
1721 int flags = vvp_io_args_flags(file, args);
1725 unsigned int retried = 0, dio_lock = 0;
1726 bool is_aio = false;
1727 bool is_parallel_dio = false;
1728 struct cl_dio_aio *ci_dio_aio = NULL;
1729 size_t per_bytes, max_io_bytes;
1734 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, bytes: %zu\n",
1735 file_dentry(file)->d_name.name,
1736 iot == CIT_READ ? "read" : "write", *ppos, bytes);
1738 max_io_bytes = min_t(size_t, PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT,
1739 sbi->ll_cache->ccc_lru_max >> 2) << PAGE_SHIFT;
1741 io = vvp_env_thread_io(env);
1742 if (iocb_ki_flags_check(flags, DIRECT)) {
1743 if (iocb_ki_flags_check(flags, APPEND))
1745 if (!is_sync_kiocb(args->u.normal.via_iocb))
1748 /* the kernel does not support AIO on pipes, and parallel DIO
1749 * uses part of the AIO path, so we must not do parallel dio
1752 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1755 if (!ll_sbi_has_parallel_dio(sbi))
1756 is_parallel_dio = false;
1758 ci_dio_aio = cl_dio_aio_alloc(args->u.normal.via_iocb,
1759 ll_i2info(inode)->lli_clob, is_aio);
1761 GOTO(out, rc = -ENOMEM);
1766 * IO block size need be aware of cached page limit, otherwise
1767 * if we have small max_cached_mb but large block IO issued, io
1768 * could not be finished and blocked whole client.
1770 if (iocb_ki_flags_check(flags, DIRECT) || bytes < max_io_bytes) {
1774 per_bytes = max_io_bytes;
1777 io = vvp_env_thread_io(env);
1778 ll_io_init(io, file, iot, args);
1779 io->ci_dio_aio = ci_dio_aio;
1780 io->ci_dio_lock = dio_lock;
1781 io->ci_ndelay_tried = retried;
1782 io->ci_parallel_dio = is_parallel_dio;
1784 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1785 if (iocb_ki_flags_check(flags, APPEND))
1786 range_lock_init(&range, 0, LUSTRE_EOF);
1788 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1790 vio->vui_fd = file->private_data;
1791 vio->vui_iter = args->u.normal.via_iter;
1792 vio->vui_iocb = args->u.normal.via_iocb;
1793 /* Direct IO reads must also take range lock,
1794 * or multiple reads will try to work on the same pages
1795 * See LU-6227 for details.
1797 if (((iot == CIT_WRITE) ||
1798 (iot == CIT_READ && iocb_ki_flags_check(flags, DIRECT))) &&
1799 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1800 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1802 rc = range_lock(&lli->lli_write_tree, &range);
1806 range_locked = true;
1809 ll_cl_add(inode, env, io, LCC_RW);
1810 rc = cl_io_loop(env, io);
1811 ll_cl_remove(inode, env);
1813 /* cl_io_rw_init() handled IO */
1817 if (io->ci_dio_aio && !is_aio) {
1818 struct cl_sync_io *anchor = &io->ci_dio_aio->cda_sync;
1820 /* for dio, EIOCBQUEUED is an implementation detail,
1821 * and we don't return it to userspace
1823 if (rc == -EIOCBQUEUED)
1826 /* N/B: parallel DIO may be disabled during i/o submission;
1827 * if that occurs, I/O shifts to sync, so it's all resolved
1828 * before we get here, and this wait call completes
1831 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1837 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1839 range_unlock(&lli->lli_write_tree, &range);
1840 range_locked = false;
1843 if (io->ci_bytes > 0) {
1845 result += io->ci_bytes;
1846 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1850 bytes -= io->ci_bytes;
1852 /* prepare IO restart */
1854 args->u.normal.via_iter = vio->vui_iter;
1858 * Reexpand iov count because it was zero
1861 iov_iter_reexpand(vio->vui_iter, bytes);
1862 if (per_bytes == io->ci_bytes)
1863 io->ci_need_restart = 1;
1867 cl_io_fini(env, io);
1870 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1871 file->f_path.dentry->d_name.name,
1872 iot, rc, result, io->ci_need_restart);
1874 if ((!rc || rc == -ENODATA || rc == -ENOLCK || rc == -EIOCBQUEUED) &&
1875 bytes > 0 && io->ci_need_restart && retries-- > 0) {
1877 "%s: restart %s from ppos=%lld bytes=%zu retries=%u ret=%zd: rc = %d\n",
1878 file_dentry(file)->d_name.name,
1879 iot == CIT_READ ? "read" : "write",
1880 *ppos, bytes, retries, result, rc);
1881 /* preserve the tried count for FLR */
1882 retried = io->ci_ndelay_tried;
1883 dio_lock = io->ci_dio_lock;
1887 if (io->ci_dio_aio) {
1888 /* set the number of bytes successfully moved in the aio */
1890 io->ci_dio_aio->cda_bytes = result;
1892 * VFS will call aio_complete() if no -EIOCBQUEUED
1893 * is returned for AIO, so we can not call aio_complete()
1894 * in our end_io(). (cda_no_aio_complete is always set for
1897 * NB: Setting cda_no_aio_complete like this is safe because
1898 * the atomic_dec_and_lock in cl_sync_io_note has implicit
1899 * memory barriers, so this will be seen by whichever thread
1900 * completes the DIO/AIO, even if it's not this one.
1902 if (is_aio && rc != -EIOCBQUEUED)
1903 io->ci_dio_aio->cda_no_aio_complete = 1;
1904 /* if an aio enqueued successfully (-EIOCBQUEUED), then Lustre
1905 * will call aio_complete rather than the vfs, so we return 0
1906 * to tell the VFS we're handling it
1908 else if (is_aio) /* rc == -EIOCBQUEUED */
1911 * Drop the reference held by the llite layer on this top level
1914 * For DIO, this frees it here, since IO is complete, and for
1915 * AIO, we will call aio_complete() (and then free this top
1916 * level context) once all the outstanding chunks of this AIO
1919 cl_sync_io_note(env, &io->ci_dio_aio->cda_sync,
1920 rc == -EIOCBQUEUED ? 0 : rc);
1922 LASSERT(io->ci_dio_aio->cda_creator_free);
1923 cl_dio_aio_free(env, io->ci_dio_aio);
1924 io->ci_dio_aio = NULL;
1928 if (iot == CIT_READ) {
1930 ll_stats_ops_tally(ll_i2sbi(inode),
1931 LPROC_LL_READ_BYTES, result);
1932 } else if (iot == CIT_WRITE) {
1934 ll_stats_ops_tally(ll_i2sbi(inode),
1935 LPROC_LL_WRITE_BYTES, result);
1936 fd->fd_write_failed = false;
1937 } else if (result == 0 && rc == 0) {
1940 fd->fd_write_failed = true;
1942 fd->fd_write_failed = false;
1943 } else if (rc != -ERESTARTSYS) {
1944 fd->fd_write_failed = true;
1948 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1950 ll_heat_add(inode, iot, result);
1952 RETURN(result > 0 ? result : rc);
1956 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1957 * especially for small I/O.
1959 * To serve a read request, CLIO has to create and initialize a cl_io and
1960 * then request DLM lock. This has turned out to have siginificant overhead
1961 * and affects the performance of small I/O dramatically.
1963 * It's not necessary to create a cl_io for each I/O. Under the help of read
1964 * ahead, most of the pages being read are already in memory cache and we can
1965 * read those pages directly because if the pages exist, the corresponding DLM
1966 * lock must exist so that page content must be valid.
1968 * In fast read implementation, the llite speculatively finds and reads pages
1969 * in memory cache. There are three scenarios for fast read:
1970 * - If the page exists and is uptodate, kernel VM will provide the data and
1971 * CLIO won't be intervened;
1972 * - If the page was brought into memory by read ahead, it will be exported
1973 * and read ahead parameters will be updated;
1974 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1975 * it will go back and invoke normal read, i.e., a cl_io will be created
1976 * and DLM lock will be requested.
1978 * POSIX compliance: posix standard states that read is intended to be atomic.
1979 * Lustre read implementation is in line with Linux kernel read implementation
1980 * and neither of them complies with POSIX standard in this matter. Fast read
1981 * doesn't make the situation worse on single node but it may interleave write
1982 * results from multiple nodes due to short read handling in ll_file_aio_read().
1984 * \param env - lu_env
1985 * \param iocb - kiocb from kernel
1986 * \param iter - user space buffers where the data will be copied
1988 * \retval - number of bytes have been read, or error code if error occurred.
1991 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1993 struct ll_inode_info *lli = ll_i2info(file_inode(iocb->ki_filp));
1994 int flags = iocb_ki_flags_get(iocb->ki_filp, iocb);
1997 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
2000 /* NB: we can't do direct IO for fast read because it will need a lock
2001 * to make IO engine happy. */
2002 if (iocb_ki_flags_check(flags, DIRECT))
2005 if (ll_layout_version_get(lli) == CL_LAYOUT_GEN_NONE)
2008 result = generic_file_read_iter(iocb, iter);
2010 /* If the first page is not in cache, generic_file_aio_read() will be
2011 * returned with -ENODATA. Fall back to full read path.
2012 * See corresponding code in ll_readpage().
2014 * if we raced with page deletion, we might get EIO. Rather than add
2015 * locking to the fast path for this rare case, fall back to the full
2016 * read path. (See vvp_io_read_start() for rest of handling.
2018 if (result == -ENODATA || result == -EIO)
2022 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
2023 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
2024 LPROC_LL_READ_BYTES, result);
2031 * Confine read iter lest read beyond the EOF
2033 * \param iocb [in] kernel iocb
2034 * \param to [in] reader iov_iter
2036 * \retval <0 failure
2038 * \retval >0 @iocb->ki_pos has passed the EOF
2040 static int file_read_confine_iter(struct lu_env *env, struct kiocb *iocb,
2041 struct iov_iter *to)
2044 struct cl_attr *attr = vvp_env_thread_attr(env);
2045 struct file *file = iocb->ki_filp;
2046 struct inode *inode = file_inode(file);
2047 struct ll_inode_info *lli = ll_i2info(inode);
2048 struct cl_object *obj = lli->lli_clob;
2049 loff_t read_end = iocb->ki_pos + iov_iter_count(to);
2059 io = vvp_env_thread_io(env);
2061 rc = cl_io_init(env, io, CIT_MISC, obj);
2065 cl_object_attr_lock(lli->lli_clob);
2066 rc = cl_object_attr_get(env, lli->lli_clob, attr);
2067 cl_object_attr_unlock(lli->lli_clob);
2070 cl_io_fini(env, io);
2074 kms = attr->cat_kms;
2075 /* if read beyond end-of-file, adjust read count */
2076 if (kms > 0 && (iocb->ki_pos >= kms || read_end > kms)) {
2077 rc = ll_glimpse_size(inode);
2081 size = i_size_read(inode);
2082 if (iocb->ki_pos >= size || read_end > size) {
2084 "%s: read [%llu, %llu] over eof, kms %llu, file_size %llu.\n",
2085 file_dentry(file)->d_name.name,
2086 iocb->ki_pos, read_end, kms, size);
2088 if (iocb->ki_pos >= size)
2091 if (read_end > size)
2092 iov_iter_truncate(to, size - iocb->ki_pos);
2100 * Read from a file (through the page cache).
2102 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2105 struct vvp_io_args *args;
2106 struct file *file = iocb->ki_filp;
2110 ktime_t kstart = ktime_get();
2112 bool stale_data = false;
2116 CDEBUG(D_VFSTRACE|D_IOTRACE,
2117 "START file %s:"DFID", ppos: %lld, count: %zu\n",
2118 file_dentry(file)->d_name.name,
2119 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2120 iov_iter_count(to));
2122 if (!iov_iter_count(to))
2125 env = cl_env_get(&refcheck);
2127 RETURN(PTR_ERR(env));
2129 result = file_read_confine_iter(env, iocb, to);
2132 else if (result > 0)
2136 * Currently when PCC read failed, we do not fall back to the
2137 * normal read path, just return the error.
2138 * The resaon is that: for RW-PCC, the file data may be modified
2139 * in the PCC and inconsistent with the data on OSTs (or file
2140 * data has been removed from the Lustre file system), at this
2141 * time, fallback to the normal read path may read the wrong
2143 * TODO: for RO-PCC (readonly PCC), fall back to normal read
2144 * path: read data from data copy on OSTs.
2146 result = pcc_file_read_iter(iocb, to, &cached);
2150 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
2152 result = ll_do_fast_read(iocb, to);
2153 if (result < 0 || iov_iter_count(to) == 0)
2156 args = ll_env_args(env);
2157 args->u.normal.via_iter = to;
2158 args->u.normal.via_iocb = iocb;
2160 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
2161 &iocb->ki_pos, iov_iter_count(to));
2164 else if (result == 0)
2168 cl_env_put(env, &refcheck);
2170 if (stale_data && result > 0) {
2172 * we've reached EOF before the read, the data read are cached
2175 iov_iter_truncate(to, 0);
2180 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2181 file->private_data, iocb->ki_pos, result,
2183 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
2184 ktime_us_delta(ktime_get(), kstart));
2188 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2189 file_dentry(file)->d_name.name,
2190 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2191 iov_iter_count(to));
2197 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
2198 * If a page is already in the page cache and dirty (and some other things -
2199 * See ll_tiny_write_begin for the instantiation of these rules), then we can
2200 * write to it without doing a full I/O, because Lustre already knows about it
2201 * and will write it out. This saves a lot of processing time.
2203 * All writes here are within one page, so exclusion is handled by the page
2204 * lock on the vm page. We do not do tiny writes for writes which touch
2205 * multiple pages because it's very unlikely multiple sequential pages are
2206 * are already dirty.
2208 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
2209 * and are unlikely to be to already dirty pages.
2211 * Attribute updates are important here, we do them in ll_tiny_write_end.
2213 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
2215 ssize_t count = iov_iter_count(iter);
2216 struct file *file = iocb->ki_filp;
2217 struct inode *inode = file_inode(file);
2218 bool lock_inode = !IS_NOSEC(inode);
2223 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2224 * of function for why.
2226 if (count >= PAGE_SIZE ||
2227 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2230 if (unlikely(lock_inode))
2231 ll_inode_lock(inode);
2232 result = __generic_file_write_iter(iocb, iter);
2234 if (unlikely(lock_inode))
2235 ll_inode_unlock(inode);
2237 /* If the page is not already dirty, ll_tiny_write_begin returns
2238 * -ENODATA. We continue on to normal write.
2240 if (result == -ENODATA)
2244 ll_heat_add(inode, CIT_WRITE, result);
2245 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2248 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2254 * Write to a file (through the page cache).
2256 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2258 struct vvp_io_args *args;
2260 ssize_t rc_tiny = 0, rc_normal;
2261 struct file *file = iocb->ki_filp;
2262 int flags = iocb_ki_flags_get(file, iocb);
2265 ktime_t kstart = ktime_get();
2270 CDEBUG(D_VFSTRACE|D_IOTRACE,
2271 "START file %s:"DFID", ppos: %lld, count: %zu\n",
2272 file_dentry(file)->d_name.name,
2273 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2274 iov_iter_count(from));
2276 if (!iov_iter_count(from))
2277 GOTO(out, rc_normal = 0);
2280 * When PCC write failed, we usually do not fall back to the normal
2281 * write path, just return the error. But there is a special case when
2282 * returned error code is -ENOSPC due to running out of space on PCC HSM
2283 * bakcend. At this time, it will fall back to normal I/O path and
2284 * retry the I/O. As the file is in HSM released state, it will restore
2285 * the file data to OSTs first and redo the write again. And the
2286 * restore process will revoke the layout lock and detach the file
2287 * from PCC cache automatically.
2289 result = pcc_file_write_iter(iocb, from, &cached);
2290 if (cached && result != -ENOSPC && result != -EDQUOT)
2291 GOTO(out, rc_normal = result);
2293 /* NB: we can't do direct IO for tiny writes because they use the page
2294 * cache, we can't do sync writes because tiny writes can't flush
2295 * pages, and we can't do append writes because we can't guarantee the
2296 * required DLM locks are held to protect file size.
2298 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2300 (ki_flag(DIRECT) | ki_flag(DSYNC) | ki_flag(SYNC) | ki_flag(APPEND))))
2301 rc_tiny = ll_do_tiny_write(iocb, from);
2303 /* In case of error, go on and try normal write - Only stop if tiny
2304 * write completed I/O.
2306 if (iov_iter_count(from) == 0)
2307 GOTO(out, rc_normal = rc_tiny);
2309 env = cl_env_get(&refcheck);
2311 RETURN(PTR_ERR(env));
2313 args = ll_env_args(env);
2314 args->u.normal.via_iter = from;
2315 args->u.normal.via_iocb = iocb;
2317 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2318 &iocb->ki_pos, iov_iter_count(from));
2320 /* On success, combine bytes written. */
2321 if (rc_tiny >= 0 && rc_normal > 0)
2322 rc_normal += rc_tiny;
2323 /* On error, only return error from normal write if tiny write did not
2324 * write any bytes. Otherwise return bytes written by tiny write.
2326 else if (rc_tiny > 0)
2327 rc_normal = rc_tiny;
2329 cl_env_put(env, &refcheck);
2331 if (rc_normal > 0) {
2332 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2333 file->private_data, iocb->ki_pos,
2335 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2336 ktime_us_delta(ktime_get(), kstart));
2340 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2341 file_dentry(file)->d_name.name,
2342 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2343 iov_iter_count(from));
2348 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2350 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2352 static int ll_file_get_iov_count(const struct iovec *iov,
2353 unsigned long *nr_segs, size_t *count,
2359 for (seg = 0; seg < *nr_segs; seg++) {
2360 const struct iovec *iv = &iov[seg];
2363 * If any segment has a negative length, or the cumulative
2364 * length ever wraps negative then return -EINVAL.
2367 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2369 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2374 cnt -= iv->iov_len; /* This segment is no good */
2381 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2382 unsigned long nr_segs, loff_t pos)
2389 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2396 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2397 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2398 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2399 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2400 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2402 result = ll_file_read_iter(iocb, &to);
2407 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2410 struct iovec iov = { .iov_base = buf, .iov_len = count };
2419 init_sync_kiocb(&kiocb, file);
2420 kiocb.ki_pos = *ppos;
2421 #ifdef HAVE_KIOCB_KI_LEFT
2422 kiocb.ki_left = count;
2423 #elif defined(HAVE_KI_NBYTES)
2424 kiocb.i_nbytes = count;
2427 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2428 *ppos = kiocb.ki_pos;
2434 * Write to a file (through the page cache).
2437 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2438 unsigned long nr_segs, loff_t pos)
2440 struct iov_iter from;
2445 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2452 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2453 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2454 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2455 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2456 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2458 result = ll_file_write_iter(iocb, &from);
2463 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2464 size_t count, loff_t *ppos)
2466 struct iovec iov = { .iov_base = (void __user *)buf,
2476 init_sync_kiocb(&kiocb, file);
2477 kiocb.ki_pos = *ppos;
2478 #ifdef HAVE_KIOCB_KI_LEFT
2479 kiocb.ki_left = count;
2480 #elif defined(HAVE_KI_NBYTES)
2481 kiocb.ki_nbytes = count;
2484 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2485 *ppos = kiocb.ki_pos;
2489 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2491 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2492 __u64 flags, struct lov_user_md *lum, int lum_size)
2494 struct lookup_intent oit = {
2496 .it_flags = flags | MDS_OPEN_BY_FID,
2501 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2502 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2503 /* this code will only exist for big-endian systems */
2504 lustre_swab_lov_user_md(lum, 0);
2507 ll_inode_size_lock(inode);
2508 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2510 GOTO(out_unlock, rc);
2512 ll_release_openhandle(dentry, &oit);
2515 ll_inode_size_unlock(inode);
2516 ll_intent_release(&oit);
2521 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2522 struct lov_mds_md **lmmp, int *lmm_size,
2523 struct ptlrpc_request **request)
2525 struct ll_sb_info *sbi = ll_i2sbi(inode);
2526 struct mdt_body *body;
2527 struct lov_mds_md *lmm = NULL;
2528 struct ptlrpc_request *req = NULL;
2529 struct md_op_data *op_data;
2534 rc = ll_get_default_mdsize(sbi, &lmmsize);
2538 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2539 strlen(filename), lmmsize,
2540 LUSTRE_OPC_ANY, NULL);
2541 if (IS_ERR(op_data))
2542 RETURN(PTR_ERR(op_data));
2544 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2545 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2546 ll_finish_md_op_data(op_data);
2548 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2553 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2554 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2556 lmmsize = body->mbo_eadatasize;
2558 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2560 GOTO(out, rc = -ENODATA);
2562 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2563 LASSERT(lmm != NULL);
2565 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2566 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2567 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2568 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2569 GOTO(out, rc = -EPROTO);
2572 * This is coming from the MDS, so is probably in
2573 * little endian. We convert it to host endian before
2574 * passing it to userspace.
2576 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2577 int stripe_count = 0;
2579 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2580 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2581 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2582 if (le32_to_cpu(lmm->lmm_pattern) &
2583 LOV_PATTERN_F_RELEASED)
2585 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2587 /* if function called for directory - we should
2588 * avoid swab not existent lsm objects
2590 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2591 S_ISREG(body->mbo_mode))
2592 lustre_swab_lov_user_md_objects(
2593 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2595 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2596 S_ISREG(body->mbo_mode))
2597 lustre_swab_lov_user_md_objects(
2598 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2600 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2601 lustre_swab_lov_comp_md_v1(
2602 (struct lov_comp_md_v1 *)lmm);
2606 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2607 struct lov_comp_md_v1 *comp_v1 = NULL;
2608 struct lov_comp_md_entry_v1 *ent;
2609 struct lov_user_md_v1 *v1 = NULL;
2613 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2614 /* Dump the striping information */
2615 for (; i < comp_v1->lcm_entry_count; i++) {
2616 ent = &comp_v1->lcm_entries[i];
2617 off = ent->lcme_offset;
2618 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2620 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2621 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2623 if (unlikely(CFS_FAIL_CHECK(OBD_FAIL_LOV_COMP_MAGIC) &&
2624 (cfs_fail_val == i + 1)))
2625 v1->lmm_magic = LOV_MAGIC_BAD;
2627 if (unlikely(CFS_FAIL_CHECK(OBD_FAIL_LOV_COMP_PATTERN) &&
2628 (cfs_fail_val == i + 1)))
2629 v1->lmm_pattern = LOV_PATTERN_BAD;
2633 GOTO(out, rc = -EINVAL);
2635 lmm->lmm_stripe_count = v1->lmm_stripe_count;
2636 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2638 * Return valid stripe_count and stripe_size instead of 0 for
2639 * DoM files to avoid divide-by-zero for older userspace that
2640 * calls this ioctl, e.g. lustre ADIO driver.
2642 if (lmm->lmm_stripe_count == 0)
2643 lmm->lmm_stripe_count = 1;
2644 if (lmm->lmm_stripe_size == 0) {
2645 /* Since the first component of the file data is placed
2646 * on the MDT for faster access, the stripe_size of the
2647 * second one is always that applications which are
2650 if (lmm->lmm_pattern & LOV_PATTERN_MDT)
2651 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2653 i = comp_v1->lcm_entry_count > 1 ?
2654 comp_v1->lcm_entry_count - 1 : 0;
2655 ent = &comp_v1->lcm_entries[i];
2656 off = ent->lcme_offset;
2657 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2658 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2663 *lmm_size = lmmsize;
2668 static int ll_lov_setea(struct inode *inode, struct file *file,
2671 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2672 struct lov_user_md *lump;
2673 int lum_size = sizeof(*lump) + sizeof(struct lov_user_ost_data);
2677 if (!capable(CAP_SYS_ADMIN))
2680 OBD_ALLOC_LARGE(lump, lum_size);
2684 if (copy_from_user(lump, arg, lum_size))
2685 GOTO(out_lump, rc = -EFAULT);
2687 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2689 cl_lov_delay_create_clear(&file->f_flags);
2692 OBD_FREE_LARGE(lump, lum_size);
2696 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2703 /* exit before doing any work if pointer is bad */
2704 if (unlikely(!ll_access_ok(lum, sizeof(struct lov_user_md))))
2707 env = cl_env_get(&refcheck);
2709 RETURN(PTR_ERR(env));
2711 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2712 cl_env_put(env, &refcheck);
2716 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2719 struct lov_user_md __user *lum = arg;
2720 struct lov_user_md *klum;
2722 __u64 flags = FMODE_WRITE;
2725 rc = ll_copy_user_md(lum, &klum);
2730 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2735 rc = put_user(0, &lum->lmm_stripe_count);
2739 rc = ll_layout_refresh(inode, &gen);
2743 rc = ll_file_getstripe(inode, arg, lum_size);
2744 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2745 ll_i2info(inode)->lli_clob) {
2746 struct iattr attr = { 0 };
2748 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2749 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2752 cl_lov_delay_create_clear(&file->f_flags);
2755 OBD_FREE_LARGE(klum, lum_size);
2761 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2763 struct ll_inode_info *lli = ll_i2info(inode);
2764 struct cl_object *obj = lli->lli_clob;
2765 struct ll_file_data *fd = file->private_data;
2766 struct ll_grouplock grouplock;
2771 CWARN("group id for group lock must not be 0\n");
2775 if (ll_file_nolock(file))
2776 RETURN(-EOPNOTSUPP);
2778 if (file->f_flags & O_NONBLOCK) {
2779 if (!mutex_trylock(&lli->lli_group_mutex))
2782 mutex_lock(&lli->lli_group_mutex);
2785 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2786 CWARN("group lock already existed with gid %lu\n",
2787 fd->fd_grouplock.lg_gid);
2788 GOTO(out, rc = -EINVAL);
2790 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2791 if (file->f_flags & O_NONBLOCK)
2792 GOTO(out, rc = -EAGAIN);
2793 mutex_unlock(&lli->lli_group_mutex);
2794 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2795 GOTO(retry, rc = 0);
2797 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2800 * XXX: group lock needs to protect all OST objects while PFL
2801 * can add new OST objects during the IO, so we'd instantiate
2802 * all OST objects before getting its group lock.
2807 struct cl_layout cl = {
2808 .cl_is_composite = false,
2810 struct lu_extent ext = {
2812 .e_end = OBD_OBJECT_EOF,
2815 env = cl_env_get(&refcheck);
2817 GOTO(out, rc = PTR_ERR(env));
2819 rc = cl_object_layout_get(env, obj, &cl);
2820 if (rc >= 0 && cl.cl_is_composite)
2821 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2824 cl_env_put(env, &refcheck);
2829 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2830 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2835 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2836 fd->fd_grouplock = grouplock;
2837 if (lli->lli_group_users == 0)
2838 lli->lli_group_gid = grouplock.lg_gid;
2839 lli->lli_group_users++;
2841 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2843 mutex_unlock(&lli->lli_group_mutex);
2848 static int ll_put_grouplock(struct inode *inode, struct file *file,
2851 struct ll_inode_info *lli = ll_i2info(inode);
2852 struct ll_file_data *fd = file->private_data;
2853 struct ll_grouplock grouplock;
2857 mutex_lock(&lli->lli_group_mutex);
2858 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2859 CWARN("no group lock held\n");
2860 GOTO(out, rc = -EINVAL);
2863 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2865 if (fd->fd_grouplock.lg_gid != arg) {
2866 CWARN("group lock %lu doesn't match current id %lu\n",
2867 arg, fd->fd_grouplock.lg_gid);
2868 GOTO(out, rc = -EINVAL);
2871 grouplock = fd->fd_grouplock;
2872 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2873 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2875 cl_put_grouplock(&grouplock);
2877 lli->lli_group_users--;
2878 if (lli->lli_group_users == 0) {
2879 lli->lli_group_gid = 0;
2880 wake_up_var(&lli->lli_group_users);
2882 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2885 mutex_unlock(&lli->lli_group_mutex);
2891 * Close inode open handle
2893 * \param dentry [in] dentry which contains the inode
2894 * \param it [in,out] intent which contains open info and result
2897 * \retval <0 failure
2899 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2901 struct inode *inode = dentry->d_inode;
2902 struct obd_client_handle *och;
2908 /* Root ? Do nothing. */
2909 if (is_root_inode(inode))
2912 /* No open handle to close? Move away */
2913 if (!it_disposition(it, DISP_OPEN_OPEN))
2916 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2918 OBD_ALLOC(och, sizeof(*och));
2920 GOTO(out, rc = -ENOMEM);
2922 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2926 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2928 /* this one is in place of ll_file_open */
2929 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2930 ptlrpc_req_finished(it->it_request);
2931 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2937 * Get size for inode for which FIEMAP mapping is requested.
2938 * Make the FIEMAP get_info call and returns the result.
2939 * \param fiemap kernel buffer to hold extens
2940 * \param num_bytes kernel buffer size
2942 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2948 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2951 /* Checks for fiemap flags */
2952 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2953 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2957 /* Check for FIEMAP_FLAG_SYNC */
2958 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2959 rc = filemap_write_and_wait(inode->i_mapping);
2964 env = cl_env_get(&refcheck);
2966 RETURN(PTR_ERR(env));
2968 if (i_size_read(inode) == 0) {
2969 rc = ll_glimpse_size(inode);
2974 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLPROJID;
2975 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2976 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2978 /* If filesize is 0, then there would be no objects for mapping */
2979 if (fmkey.lfik_oa.o_size == 0) {
2980 fiemap->fm_mapped_extents = 0;
2984 fmkey.lfik_fiemap = *fiemap;
2986 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2987 &fmkey, fiemap, &num_bytes);
2989 cl_env_put(env, &refcheck);
2993 static int fid2path_for_enc_file(struct inode *parent, char *gfpath,
2996 struct dentry *de = NULL, *de_parent = d_find_any_alias(parent);
2997 struct llcrypt_str lltr = LLTR_INIT(NULL, 0);
2998 struct llcrypt_str de_name;
2999 char *p, *ptr = gfpath;
3000 size_t len = 0, len_orig = 0;
3001 int enckey = -1, nameenc = -1;
3005 while ((p = strsep(&gfpath, "/")) != NULL) {
3013 len_orig = strlen(p);
3015 rc = sscanf(p, "["SFID"]", RFID(&fid));
3017 p = strchr(p, ']') + 1;
3023 if (!IS_ENCRYPTED(parent)) {
3024 if (gfpathlen < len + 1) {
3029 memmove(ptr, p, len);
3033 gfpathlen -= len + 1;
3037 /* From here, we know parent is encrypted */
3040 rc = llcrypt_prepare_readdir(parent);
3041 if (rc && rc != -ENOKEY) {
3048 if (llcrypt_has_encryption_key(parent))
3054 llcrypt_policy_has_filename_enc(parent);
3057 /* Even if names are not encrypted, we still need to call
3058 * ll_fname_disk_to_usr in order to decode names as they are
3059 * coming from the wire.
3061 rc = llcrypt_fname_alloc_buffer(parent, NAME_MAX + 1, &lltr);
3069 rc = ll_fname_disk_to_usr(parent, 0, 0, &de_name,
3072 llcrypt_fname_free_buffer(&lltr);
3076 lltr.name[lltr.len] = '\0';
3078 if (lltr.len <= len_orig && gfpathlen >= lltr.len + 1) {
3079 memcpy(ptr, lltr.name, lltr.len);
3084 gfpathlen -= lltr.len + 1;
3088 llcrypt_fname_free_buffer(&lltr);
3090 if (rc == -EOVERFLOW) {
3097 /* We reached the end of the string, which means
3098 * we are dealing with the last component in the path.
3099 * So save a useless lookup and exit.
3105 if (enckey == 0 || nameenc == 0)
3108 ll_inode_lock(parent);
3109 de = lookup_one_len(p, de_parent, len);
3110 ll_inode_unlock(parent);
3111 if (IS_ERR_OR_NULL(de) || !de->d_inode) {
3117 parent = de->d_inode;
3124 if (!IS_ERR_OR_NULL(de))
3129 int __ll_fid2path(struct inode *inode, struct getinfo_fid2path *gfout,
3130 size_t outsize, __u32 pathlen_orig)
3132 struct obd_export *exp = ll_i2mdexp(inode);
3135 /* Append root FID after gfout to let MDT know the root FID so that
3136 * it can lookup the correct path, this is mainly for fileset.
3137 * old server without fileset mount support will ignore this.
3139 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
3141 /* Call mdc_iocontrol */
3142 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
3144 if (!rc && gfout->gf_pathlen && gfout->gf_u.gf_path[0] == '/') {
3145 /* by convention, server side (mdt_path_current()) puts
3146 * a leading '/' to tell client that we are dealing with
3149 rc = fid2path_for_enc_file(inode, gfout->gf_u.gf_path,
3151 if (!rc && strlen(gfout->gf_u.gf_path) > pathlen_orig)
3158 int ll_fid2path(struct inode *inode, void __user *arg)
3160 const struct getinfo_fid2path __user *gfin = arg;
3161 __u32 pathlen, pathlen_orig;
3162 struct getinfo_fid2path *gfout;
3168 if (!capable(CAP_DAC_READ_SEARCH) &&
3169 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3172 /* Only need to get the buflen */
3173 if (get_user(pathlen, &gfin->gf_pathlen))
3176 if (pathlen > PATH_MAX)
3178 pathlen_orig = pathlen;
3181 outsize = sizeof(*gfout) + pathlen;
3182 OBD_ALLOC(gfout, outsize);
3186 if (copy_from_user(gfout, arg, sizeof(*gfout)))
3187 GOTO(gf_free, rc = -EFAULT);
3189 gfout->gf_pathlen = pathlen;
3190 rc = __ll_fid2path(inode, gfout, outsize, pathlen_orig);
3194 if (copy_to_user(arg, gfout, sizeof(*gfout) + pathlen_orig))
3198 OBD_FREE(gfout, outsize);
3199 if (rc == -ENAMETOOLONG) {
3200 pathlen += PATH_MAX;
3207 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
3209 struct cl_object *obj = ll_i2info(inode)->lli_clob;
3217 ioc->idv_version = 0;
3218 ioc->idv_layout_version = UINT_MAX;
3220 /* If no file object initialized, we consider its version is 0. */
3224 env = cl_env_get(&refcheck);
3226 RETURN(PTR_ERR(env));
3228 io = vvp_env_thread_io(env);
3230 io->u.ci_data_version.dv_data_version = 0;
3231 io->u.ci_data_version.dv_layout_version = UINT_MAX;
3232 io->u.ci_data_version.dv_flags = ioc->idv_flags;
3235 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
3236 result = cl_io_loop(env, io);
3238 result = io->ci_result;
3240 ioc->idv_version = io->u.ci_data_version.dv_data_version;
3241 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
3242 cl_io_fini(env, io);
3244 if (unlikely(io->ci_need_restart))
3247 cl_env_put(env, &refcheck);
3253 * Read the data_version for inode.
3255 * This value is computed using stripe object version on OST.
3256 * Version is computed using server side locking.
3258 * @param flags if do sync on the OST side;
3260 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
3261 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
3263 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
3265 struct ioc_data_version ioc = { .idv_flags = flags };
3268 rc = ll_ioc_data_version(inode, &ioc);
3270 *data_version = ioc.idv_version;
3276 * Trigger a HSM release request for the provided inode.
3278 int ll_hsm_release(struct inode *inode)
3281 struct obd_client_handle *och = NULL;
3282 __u64 data_version = 0;
3288 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
3289 ll_i2sbi(inode)->ll_fsname,
3290 PFID(&ll_i2info(inode)->lli_fid));
3292 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
3294 GOTO(out, rc = PTR_ERR(och));
3296 /* Grab latest data_version and [am]time values */
3297 rc = ll_data_version(inode, &data_version,
3298 LL_DV_WR_FLUSH | LL_DV_SZ_UPDATE);
3302 env = cl_env_get(&refcheck);
3304 GOTO(out, rc = PTR_ERR(env));
3306 rc = ll_merge_attr(env, inode);
3307 cl_env_put(env, &refcheck);
3309 /* If error happen, we have the wrong size for a file.
3315 /* Release the file.
3316 * NB: lease lock handle is released in mdc_hsm_release_pack() because
3317 * we still need it to pack l_remote_handle to MDT. */
3318 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
3324 if (och != NULL && !IS_ERR(och)) /* close the file */
3325 ll_lease_close(och, inode, NULL);
3330 struct ll_swap_stack {
3333 struct inode *inode1;
3334 struct inode *inode2;
3339 static int ll_swap_layouts(struct file *file1, struct file *file2,
3340 struct lustre_swap_layouts *lsl)
3342 struct mdc_swap_layouts msl;
3343 struct md_op_data *op_data;
3346 struct ll_swap_stack *llss = NULL;
3349 OBD_ALLOC_PTR(llss);
3353 llss->inode1 = file_inode(file1);
3354 llss->inode2 = file_inode(file2);
3356 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
3360 /* we use 2 bool because it is easier to swap than 2 bits */
3361 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
3362 llss->check_dv1 = true;
3364 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
3365 llss->check_dv2 = true;
3367 /* we cannot use lsl->sl_dvX directly because we may swap them */
3368 llss->dv1 = lsl->sl_dv1;
3369 llss->dv2 = lsl->sl_dv2;
3371 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
3372 if (rc == 0) /* same file, done! */
3375 if (rc < 0) { /* sequentialize it */
3376 swap(llss->inode1, llss->inode2);
3378 swap(llss->dv1, llss->dv2);
3379 swap(llss->check_dv1, llss->check_dv2);
3383 if (gid != 0) { /* application asks to flush dirty cache */
3384 rc = ll_get_grouplock(llss->inode1, file1, gid);
3388 rc = ll_get_grouplock(llss->inode2, file2, gid);
3390 ll_put_grouplock(llss->inode1, file1, gid);
3395 /* ultimate check, before swaping the layouts we check if
3396 * dataversion has changed (if requested) */
3397 if (llss->check_dv1) {
3398 rc = ll_data_version(llss->inode1, &dv, 0);
3401 if (dv != llss->dv1)
3402 GOTO(putgl, rc = -EAGAIN);
3405 if (llss->check_dv2) {
3406 rc = ll_data_version(llss->inode2, &dv, 0);
3409 if (dv != llss->dv2)
3410 GOTO(putgl, rc = -EAGAIN);
3413 /* struct md_op_data is used to send the swap args to the mdt
3414 * only flags is missing, so we use struct mdc_swap_layouts
3415 * through the md_op_data->op_data */
3416 /* flags from user space have to be converted before they are send to
3417 * server, no flag is sent today, they are only used on the client */
3420 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3421 0, LUSTRE_OPC_ANY, &msl);
3422 if (IS_ERR(op_data))
3423 GOTO(free, rc = PTR_ERR(op_data));
3425 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3426 sizeof(*op_data), op_data, NULL);
3427 ll_finish_md_op_data(op_data);
3434 ll_put_grouplock(llss->inode2, file2, gid);
3435 ll_put_grouplock(llss->inode1, file1, gid);
3445 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3447 struct obd_export *exp = ll_i2mdexp(inode);
3448 struct md_op_data *op_data;
3452 /* Detect out-of range masks */
3453 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3456 /* Non-root users are forbidden to set or clear flags which are
3457 * NOT defined in HSM_USER_MASK. */
3458 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3459 !capable(CAP_SYS_ADMIN))
3462 if (!exp_connect_archive_id_array(exp)) {
3463 /* Detect out-of range archive id */
3464 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3465 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3469 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3470 LUSTRE_OPC_ANY, hss);
3471 if (IS_ERR(op_data))
3472 RETURN(PTR_ERR(op_data));
3474 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3477 ll_finish_md_op_data(op_data);
3482 static int ll_hsm_import(struct inode *inode, struct file *file,
3483 struct hsm_user_import *hui)
3485 struct hsm_state_set *hss = NULL;
3486 struct iattr *attr = NULL;
3490 if (!S_ISREG(inode->i_mode))
3496 GOTO(out, rc = -ENOMEM);
3498 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3499 hss->hss_archive_id = hui->hui_archive_id;
3500 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3501 rc = ll_hsm_state_set(inode, hss);
3505 OBD_ALLOC_PTR(attr);
3507 GOTO(out, rc = -ENOMEM);
3509 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3510 attr->ia_mode |= S_IFREG;
3511 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3512 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3513 attr->ia_size = hui->hui_size;
3514 attr->ia_mtime.tv_sec = hui->hui_mtime;
3515 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3516 attr->ia_atime.tv_sec = hui->hui_atime;
3517 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3519 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3520 ATTR_UID | ATTR_GID |
3521 ATTR_MTIME | ATTR_MTIME_SET |
3522 ATTR_ATIME | ATTR_ATIME_SET;
3525 /* inode lock owner set in ll_setattr_raw()*/
3526 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3529 inode_unlock(inode);
3541 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3543 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3544 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3547 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3549 struct inode *inode = file_inode(file);
3551 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3552 ATTR_MTIME | ATTR_MTIME_SET |
3555 .tv_sec = lfu->lfu_atime_sec,
3556 .tv_nsec = lfu->lfu_atime_nsec,
3559 .tv_sec = lfu->lfu_mtime_sec,
3560 .tv_nsec = lfu->lfu_mtime_nsec,
3563 .tv_sec = lfu->lfu_ctime_sec,
3564 .tv_nsec = lfu->lfu_ctime_nsec,
3570 if (!capable(CAP_SYS_ADMIN))
3573 if (!S_ISREG(inode->i_mode))
3577 /* inode lock owner set in ll_setattr_raw()*/
3578 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3580 inode_unlock(inode);
3585 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3588 case MODE_READ_USER:
3590 case MODE_WRITE_USER:
3597 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3599 /* Used to allow the upper layers of the client to request an LDLM lock
3600 * without doing an actual read or write.
3602 * Used for ladvise lockahead to manually request specific locks.
3604 * \param[in] file file this ladvise lock request is on
3605 * \param[in] ladvise ladvise struct describing this lock request
3607 * \retval 0 success, no detailed result available (sync requests
3608 * and requests sent to the server [not handled locally]
3609 * cannot return detailed results)
3610 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3611 * see definitions for details.
3612 * \retval negative negative errno on error
3614 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3616 struct lu_env *env = NULL;
3617 struct cl_io *io = NULL;
3618 struct cl_lock *lock = NULL;
3619 struct cl_lock_descr *descr = NULL;
3620 struct dentry *dentry = file->f_path.dentry;
3621 struct inode *inode = dentry->d_inode;
3622 enum cl_lock_mode cl_mode;
3623 off_t start = ladvise->lla_start;
3624 off_t end = ladvise->lla_end;
3631 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3632 dentry, dentry->d_inode,
3633 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3636 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3638 GOTO(out, result = cl_mode);
3640 /* Get IO environment */
3641 result = cl_io_get(inode, &env, &io, &refcheck);
3645 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3648 * nothing to do for this io. This currently happens when
3649 * stripe sub-object's are not yet created.
3651 result = io->ci_result;
3652 } else if (result == 0) {
3653 lock = vvp_env_lock(env);
3654 descr = &lock->cll_descr;
3656 descr->cld_obj = io->ci_obj;
3657 /* Convert byte offsets to pages */
3658 descr->cld_start = start >> PAGE_SHIFT;
3659 descr->cld_end = end >> PAGE_SHIFT;
3660 descr->cld_mode = cl_mode;
3661 /* CEF_MUST is used because we do not want to convert a
3662 * lockahead request to a lockless lock */
3663 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3665 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3666 descr->cld_enq_flags |= CEF_SPECULATIVE;
3668 result = cl_lock_request(env, io, lock);
3670 /* On success, we need to release the lock */
3672 cl_lock_release(env, lock);
3674 cl_io_fini(env, io);
3675 cl_env_put(env, &refcheck);
3677 /* -ECANCELED indicates a matching lock with a different extent
3678 * was already present, and -EEXIST indicates a matching lock
3679 * on exactly the same extent was already present.
3680 * We convert them to positive values for userspace to make
3681 * recognizing true errors easier.
3682 * Note we can only return these detailed results on async requests,
3683 * as sync requests look the same as i/o requests for locking. */
3684 if (result == -ECANCELED)
3685 result = LLA_RESULT_DIFFERENT;
3686 else if (result == -EEXIST)
3687 result = LLA_RESULT_SAME;
3692 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3694 static int ll_ladvise_sanity(struct inode *inode,
3695 struct llapi_lu_ladvise *ladvise)
3697 struct ll_sb_info *sbi = ll_i2sbi(inode);
3698 enum lu_ladvise_type advice = ladvise->lla_advice;
3699 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3700 * be in the first 32 bits of enum ladvise_flags */
3701 __u32 flags = ladvise->lla_peradvice_flags;
3702 /* 3 lines at 80 characters per line, should be plenty */
3705 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3708 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3709 sbi->ll_fsname, advice,
3710 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3714 /* Per-advice checks */
3716 case LU_LADVISE_LOCKNOEXPAND:
3717 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3719 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3720 "rc = %d\n", sbi->ll_fsname, flags,
3721 ladvise_names[advice], rc);
3725 case LU_LADVISE_LOCKAHEAD:
3726 /* Currently only READ and WRITE modes can be requested */
3727 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3728 ladvise->lla_lockahead_mode == 0) {
3730 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3731 "rc = %d\n", sbi->ll_fsname,
3732 ladvise->lla_lockahead_mode,
3733 ladvise_names[advice], rc);
3737 case LU_LADVISE_WILLREAD:
3738 case LU_LADVISE_DONTNEED:
3740 /* Note fall through above - These checks apply to all advices
3741 * except LOCKNOEXPAND */
3742 if (flags & ~LF_DEFAULT_MASK) {
3744 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3745 "rc = %d\n", sbi->ll_fsname, flags,
3746 ladvise_names[advice], rc);
3749 if (ladvise->lla_start >= ladvise->lla_end) {
3751 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3752 "for %s: rc = %d\n", sbi->ll_fsname,
3753 ladvise->lla_start, ladvise->lla_end,
3754 ladvise_names[advice], rc);
3766 * Give file access advices
3768 * The ladvise interface is similar to Linux fadvise() system call, except it
3769 * forwards the advices directly from Lustre client to server. The server side
3770 * codes will apply appropriate read-ahead and caching techniques for the
3771 * corresponding files.
3773 * A typical workload for ladvise is e.g. a bunch of different clients are
3774 * doing small random reads of a file, so prefetching pages into OSS cache
3775 * with big linear reads before the random IO is a net benefit. Fetching
3776 * all that data into each client cache with fadvise() may not be, due to
3777 * much more data being sent to the client.
3779 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3780 struct llapi_lu_ladvise *ladvise)
3784 struct cl_ladvise_io *lio;
3789 env = cl_env_get(&refcheck);
3791 RETURN(PTR_ERR(env));
3793 io = vvp_env_thread_io(env);
3794 io->ci_obj = ll_i2info(inode)->lli_clob;
3796 /* initialize parameters for ladvise */
3797 lio = &io->u.ci_ladvise;
3798 lio->lio_start = ladvise->lla_start;
3799 lio->lio_end = ladvise->lla_end;
3800 lio->lio_fid = ll_inode2fid(inode);
3801 lio->lio_advice = ladvise->lla_advice;
3802 lio->lio_flags = flags;
3804 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3805 rc = cl_io_loop(env, io);
3809 cl_io_fini(env, io);
3810 cl_env_put(env, &refcheck);
3814 static int ll_lock_noexpand(struct file *file, int flags)
3816 struct ll_file_data *fd = file->private_data;
3818 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3823 #ifndef HAVE_FILEATTR_GET
3824 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3827 struct fsxattr fsxattr;
3829 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3832 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3833 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3834 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3835 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3836 if (copy_to_user(uarg, &fsxattr, sizeof(fsxattr)))
3843 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3847 * Project Quota ID state is only allowed to change from within the init
3848 * namespace. Enforce that restriction only if we are trying to change
3849 * the quota ID state. Everything else is allowed in user namespaces.
3851 if (current_user_ns() == &init_user_ns) {
3853 * Caller is allowed to change the project ID. if it is being
3854 * changed, make sure that the new value is valid.
3856 if (ll_i2info(inode)->lli_projid != projid &&
3857 !projid_valid(make_kprojid(&init_user_ns, projid)))
3863 if (ll_i2info(inode)->lli_projid != projid)
3866 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3867 if (!(xflags & FS_XFLAG_PROJINHERIT))
3870 if (xflags & FS_XFLAG_PROJINHERIT)
3877 int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3879 struct ptlrpc_request *req = NULL;
3880 struct md_op_data *op_data;
3881 struct cl_object *obj;
3882 unsigned int inode_flags;
3885 CDEBUG(D_QUOTA, DFID" xflags=%x projid=%u\n",
3886 PFID(ll_inode2fid(inode)), xflags, projid);
3887 rc = ll_ioctl_check_project(inode, xflags, projid);
3891 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3892 LUSTRE_OPC_ANY, NULL);
3893 if (IS_ERR(op_data))
3894 RETURN(PTR_ERR(op_data));
3896 inode_flags = ll_xflags_to_inode_flags(xflags);
3897 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3898 if (xflags & FS_XFLAG_PROJINHERIT)
3899 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3901 /* pass projid to md_op_data */
3902 op_data->op_projid = projid;
3904 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3905 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3906 ptlrpc_req_finished(req);
3908 GOTO(out_fsxattr, rc);
3909 ll_update_inode_flags(inode, op_data->op_attr_flags);
3911 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3912 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3913 GOTO(out_fsxattr, rc);
3915 obj = ll_i2info(inode)->lli_clob;
3917 struct iattr attr = { 0 };
3919 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3923 ll_finish_md_op_data(op_data);
3927 #ifndef HAVE_FILEATTR_GET
3928 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3931 struct fsxattr fsxattr;
3935 if (copy_from_user(&fsxattr, uarg, sizeof(fsxattr)))
3938 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3939 fsxattr.fsx_projid));
3943 int ll_ioctl_project(struct file *file, unsigned int cmd, void __user *uarg)
3945 struct lu_project lu_project;
3946 struct dentry *dentry = file_dentry(file);
3947 struct inode *inode = file_inode(file);
3948 struct dentry *child_dentry = NULL;
3949 int rc = 0, name_len;
3951 if (copy_from_user(&lu_project, uarg, sizeof(lu_project)))
3954 /* apply child dentry if name is valid */
3955 name_len = strnlen(lu_project.project_name, NAME_MAX);
3956 if (name_len > 0 && name_len <= NAME_MAX) {
3957 ll_inode_lock(inode);
3958 child_dentry = lookup_one_len(lu_project.project_name,
3960 ll_inode_unlock(inode);
3961 if (IS_ERR(child_dentry)) {
3962 rc = PTR_ERR(child_dentry);
3965 inode = child_dentry->d_inode;
3970 } else if (name_len > NAME_MAX) {
3975 switch (lu_project.project_type) {
3976 case LU_PROJECT_SET:
3977 rc = ll_set_project(inode, lu_project.project_xflags,
3978 lu_project.project_id);
3980 case LU_PROJECT_GET:
3981 lu_project.project_xflags =
3982 ll_inode_flags_to_xflags(inode->i_flags);
3983 if (test_bit(LLIF_PROJECT_INHERIT,
3984 &ll_i2info(inode)->lli_flags))
3985 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3986 lu_project.project_id = ll_i2info(inode)->lli_projid;
3987 if (copy_to_user(uarg, &lu_project, sizeof(lu_project))) {
3997 if (!IS_ERR_OR_NULL(child_dentry))
4002 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
4005 struct inode *inode = file_inode(file);
4006 struct ll_file_data *fd = file->private_data;
4007 struct ll_inode_info *lli = ll_i2info(inode);
4008 struct obd_client_handle *och = NULL;
4009 struct split_param sp;
4010 struct pcc_param param;
4011 bool lease_broken = false;
4013 enum mds_op_bias bias = 0;
4015 struct file *layout_file = NULL;
4017 size_t data_size = 0;
4018 bool attached = false;
4023 mutex_lock(&lli->lli_och_mutex);
4024 if (fd->fd_lease_och != NULL) {
4025 och = fd->fd_lease_och;
4026 fd->fd_lease_och = NULL;
4028 mutex_unlock(&lli->lli_och_mutex);
4033 fmode = och->och_flags;
4035 switch (ioc->lil_flags) {
4036 case LL_LEASE_RESYNC_DONE:
4037 if (ioc->lil_count > IOC_IDS_MAX)
4038 GOTO(out_lease_close, rc = -EINVAL);
4040 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
4041 OBD_ALLOC(data, data_size);
4043 GOTO(out_lease_close, rc = -ENOMEM);
4045 if (copy_from_user(data, uarg, data_size))
4046 GOTO(out_lease_close, rc = -EFAULT);
4048 bias = MDS_CLOSE_RESYNC_DONE;
4050 case LL_LEASE_LAYOUT_MERGE:
4051 if (ioc->lil_count != 1)
4052 GOTO(out_lease_close, rc = -EINVAL);
4054 uarg += sizeof(*ioc);
4055 if (copy_from_user(&fdv, uarg, sizeof(fdv)))
4056 GOTO(out_lease_close, rc = -EFAULT);
4058 layout_file = fget(fdv);
4060 GOTO(out_lease_close, rc = -EBADF);
4062 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
4063 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
4064 GOTO(out_lease_close, rc = -EPERM);
4066 data = file_inode(layout_file);
4067 bias = MDS_CLOSE_LAYOUT_MERGE;
4069 case LL_LEASE_LAYOUT_SPLIT: {
4072 if (ioc->lil_count != 2)
4073 GOTO(out_lease_close, rc = -EINVAL);
4075 uarg += sizeof(*ioc);
4076 if (copy_from_user(&fdv, uarg, sizeof(fdv)))
4077 GOTO(out_lease_close, rc = -EFAULT);
4079 uarg += sizeof(fdv);
4080 if (copy_from_user(&mirror_id, uarg, sizeof(mirror_id)))
4081 GOTO(out_lease_close, rc = -EFAULT);
4082 if (mirror_id >= MIRROR_ID_NEG)
4083 GOTO(out_lease_close, rc = -EINVAL);
4085 layout_file = fget(fdv);
4087 GOTO(out_lease_close, rc = -EBADF);
4089 /* if layout_file == file, it means to destroy the mirror */
4090 sp.sp_inode = file_inode(layout_file);
4091 sp.sp_mirror_id = (__u16)mirror_id;
4093 bias = MDS_CLOSE_LAYOUT_SPLIT;
4096 case LL_LEASE_PCC_ATTACH:
4097 if (ioc->lil_count != 1)
4100 if (IS_ENCRYPTED(inode))
4101 RETURN(-EOPNOTSUPP);
4103 uarg += sizeof(*ioc);
4104 if (copy_from_user(¶m.pa_archive_id, uarg, sizeof(__u32)))
4105 GOTO(out_lease_close, rc2 = -EFAULT);
4107 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
4109 GOTO(out_lease_close, rc2);
4112 /* Grab latest data version */
4113 rc2 = ll_data_version(inode, ¶m.pa_data_version,
4116 GOTO(out_lease_close, rc2);
4119 bias = MDS_PCC_ATTACH;
4122 /* without close intent */
4127 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
4131 rc = ll_lease_och_release(inode, file);
4140 if (ioc->lil_flags == LL_LEASE_RESYNC_DONE && data)
4141 OBD_FREE(data, data_size);
4146 if (ioc->lil_flags == LL_LEASE_PCC_ATTACH) {
4149 rc = pcc_readwrite_attach_fini(file, inode,
4150 param.pa_layout_gen,
4155 ll_layout_refresh(inode, &fd->fd_layout_version);
4158 rc = ll_lease_type_from_fmode(fmode);
4162 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
4165 struct inode *inode = file_inode(file);
4166 struct ll_inode_info *lli = ll_i2info(inode);
4167 struct ll_file_data *fd = file->private_data;
4168 struct obd_client_handle *och = NULL;
4169 __u64 open_flags = 0;
4175 switch (ioc->lil_mode) {
4176 case LL_LEASE_WRLCK:
4177 if (!(file->f_mode & FMODE_WRITE))
4179 fmode = FMODE_WRITE;
4181 case LL_LEASE_RDLCK:
4182 if (!(file->f_mode & FMODE_READ))
4186 case LL_LEASE_UNLCK:
4187 RETURN(ll_file_unlock_lease(file, ioc, uarg));
4192 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
4194 /* apply for lease */
4195 if (ioc->lil_flags & LL_LEASE_RESYNC)
4196 open_flags = MDS_OPEN_RESYNC;
4197 och = ll_lease_open(inode, file, fmode, open_flags);
4199 RETURN(PTR_ERR(och));
4201 if (ioc->lil_flags & LL_LEASE_RESYNC) {
4202 rc = ll_lease_file_resync(och, inode, uarg);
4204 ll_lease_close(och, inode, NULL);
4207 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
4209 ll_lease_close(och, inode, NULL);
4215 mutex_lock(&lli->lli_och_mutex);
4216 if (fd->fd_lease_och == NULL) {
4217 fd->fd_lease_och = och;
4220 mutex_unlock(&lli->lli_och_mutex);
4222 /* impossible now that only excl is supported for now */
4223 ll_lease_close(och, inode, &lease_broken);
4229 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
4231 struct ll_inode_info *lli = ll_i2info(inode);
4232 struct ll_sb_info *sbi = ll_i2sbi(inode);
4233 __u64 now = ktime_get_real_seconds();
4236 spin_lock(&lli->lli_heat_lock);
4237 heat->lh_flags = lli->lli_heat_flags;
4238 for (i = 0; i < heat->lh_count; i++)
4239 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
4240 now, sbi->ll_heat_decay_weight,
4241 sbi->ll_heat_period_second);
4242 spin_unlock(&lli->lli_heat_lock);
4245 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
4247 struct ll_inode_info *lli = ll_i2info(inode);
4250 spin_lock(&lli->lli_heat_lock);
4251 if (flags & LU_HEAT_FLAG_CLEAR)
4252 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
4254 if (flags & LU_HEAT_FLAG_OFF)
4255 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
4257 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
4259 spin_unlock(&lli->lli_heat_lock);
4265 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4267 struct inode *inode = file_inode(file);
4268 struct ll_file_data *fd = file->private_data;
4269 void __user *uarg = (void __user *)arg;
4273 CDEBUG(D_VFSTRACE|D_IOCTL, "VFS Op:inode="DFID"(%pK) cmd=%x arg=%lx\n",
4274 PFID(ll_inode2fid(inode)), inode, cmd, arg);
4275 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
4277 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
4278 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
4281 /* can't do a generic karg == NULL check here, since it is too noisy and
4282 * we need to return -ENOTTY for unsupported ioctls instead of -EINVAL.
4285 case LL_IOC_GETFLAGS:
4286 /* Get the current value of the file flags */
4287 return put_user(fd->fd_flags, (int __user *)arg);
4288 case LL_IOC_SETFLAGS:
4289 case LL_IOC_CLRFLAGS:
4290 /* Set or clear specific file flags */
4291 /* XXX This probably needs checks to ensure the flags are
4292 * not abused, and to handle any flag side effects.
4294 if (get_user(flags, (int __user *)arg))
4297 if (cmd == LL_IOC_SETFLAGS) {
4298 if ((flags & LL_FILE_IGNORE_LOCK) &&
4299 !(file->f_flags & O_DIRECT)) {
4301 CERROR("%s: unable to disable locking on non-O_DIRECT file "DFID": rc = %d\n",
4302 current->comm, PFID(ll_inode2fid(inode)),
4307 fd->fd_flags |= flags;
4309 fd->fd_flags &= ~flags;
4312 case LL_IOC_LOV_SETSTRIPE:
4313 case LL_IOC_LOV_SETSTRIPE_NEW:
4314 RETURN(ll_lov_setstripe(inode, file, uarg));
4315 case LL_IOC_LOV_SETEA:
4316 RETURN(ll_lov_setea(inode, file, uarg));
4317 case LL_IOC_LOV_SWAP_LAYOUTS: {
4319 struct lustre_swap_layouts lsl;
4321 if (copy_from_user(&lsl, uarg, sizeof(lsl)))
4324 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
4327 file2 = fget(lsl.sl_fd);
4331 /* O_WRONLY or O_RDWR */
4332 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
4333 GOTO(out, rc = -EPERM);
4335 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
4336 struct obd_client_handle *och = NULL;
4337 struct ll_inode_info *lli;
4338 struct inode *inode2;
4340 lli = ll_i2info(inode);
4341 mutex_lock(&lli->lli_och_mutex);
4342 if (fd->fd_lease_och != NULL) {
4343 och = fd->fd_lease_och;
4344 fd->fd_lease_och = NULL;
4346 mutex_unlock(&lli->lli_och_mutex);
4348 GOTO(out, rc = -ENOLCK);
4349 inode2 = file_inode(file2);
4350 rc = ll_swap_layouts_close(och, inode, inode2);
4352 rc = ll_swap_layouts(file, file2, &lsl);
4358 case LL_IOC_LOV_GETSTRIPE:
4359 case LL_IOC_LOV_GETSTRIPE_NEW:
4360 RETURN(ll_file_getstripe(inode, uarg, 0));
4361 case LL_IOC_GROUP_LOCK:
4362 RETURN(ll_get_grouplock(inode, file, arg));
4363 case LL_IOC_GROUP_UNLOCK:
4364 RETURN(ll_put_grouplock(inode, file, arg));
4365 case LL_IOC_DATA_VERSION: {
4366 struct ioc_data_version idv;
4369 if (copy_from_user(&idv, uarg, sizeof(idv)))
4372 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
4373 rc = ll_ioc_data_version(inode, &idv);
4375 if (rc == 0 && copy_to_user(uarg, &idv, sizeof(idv)))
4380 case LL_IOC_HSM_STATE_GET: {
4381 struct md_op_data *op_data;
4382 struct hsm_user_state *hus;
4385 if (!ll_access_ok(uarg, sizeof(*hus)))
4392 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4393 LUSTRE_OPC_ANY, hus);
4394 if (IS_ERR(op_data)) {
4395 rc = PTR_ERR(op_data);
4397 rc = obd_iocontrol(cmd, ll_i2mdexp(inode),
4398 sizeof(*op_data), op_data, NULL);
4400 if (copy_to_user(uarg, hus, sizeof(*hus)))
4403 ll_finish_md_op_data(op_data);
4408 case LL_IOC_HSM_STATE_SET: {
4409 struct hsm_state_set *hss;
4416 if (copy_from_user(hss, uarg, sizeof(*hss)))
4419 rc = ll_hsm_state_set(inode, hss);
4424 case LL_IOC_HSM_ACTION: {
4425 struct md_op_data *op_data;
4426 struct hsm_current_action *hca;
4430 if (!ll_access_ok(uarg, sizeof(*hca)))
4437 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4438 LUSTRE_OPC_ANY, hca);
4439 if (IS_ERR(op_data)) {
4441 RETURN(PTR_ERR(op_data));
4444 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4447 GOTO(skip_copy, rc);
4449 /* The hsm_current_action retreived from the server could
4450 * contain corrupt information. If it is incorrect data collect
4451 * debug information. We still send the data even if incorrect
4452 * to user land to handle.
4454 action = hsm_user_action2name(hca->hca_action);
4455 if (strcmp(action, "UNKNOWN") == 0 ||
4456 hca->hca_state > HPS_DONE) {
4458 "HSM current state %s action %s, offset = %llu, length %llu\n",
4459 hsm_progress_state2name(hca->hca_state), action,
4460 hca->hca_location.offset, hca->hca_location.length);
4463 if (copy_to_user(uarg, hca, sizeof(*hca)))
4466 ll_finish_md_op_data(op_data);
4470 case LL_IOC_SET_LEASE_OLD: {
4471 struct ll_ioc_lease ioc = { .lil_mode = arg };
4473 RETURN(ll_file_set_lease(file, &ioc, 0));
4475 case LL_IOC_SET_LEASE: {
4476 struct ll_ioc_lease ioc;
4478 if (copy_from_user(&ioc, uarg, sizeof(ioc)))
4481 RETURN(ll_file_set_lease(file, &ioc, uarg));
4483 case LL_IOC_GET_LEASE: {
4484 struct ll_inode_info *lli = ll_i2info(inode);
4485 struct ldlm_lock *lock = NULL;
4488 mutex_lock(&lli->lli_och_mutex);
4489 if (fd->fd_lease_och != NULL) {
4490 struct obd_client_handle *och = fd->fd_lease_och;
4492 lock = ldlm_handle2lock(&och->och_lease_handle);
4494 lock_res_and_lock(lock);
4495 if (!ldlm_is_cancel(lock))
4496 fmode = och->och_flags;
4498 unlock_res_and_lock(lock);
4499 LDLM_LOCK_PUT(lock);
4502 mutex_unlock(&lli->lli_och_mutex);
4504 RETURN(ll_lease_type_from_fmode(fmode));
4506 case LL_IOC_HSM_IMPORT: {
4507 struct hsm_user_import *hui;
4513 if (copy_from_user(hui, uarg, sizeof(*hui)))
4516 rc = ll_hsm_import(inode, file, hui);
4521 case LL_IOC_FUTIMES_3: {
4522 struct ll_futimes_3 lfu;
4524 if (copy_from_user(&lfu, uarg, sizeof(lfu)))
4527 RETURN(ll_file_futimes_3(file, &lfu));
4529 case LL_IOC_LADVISE: {
4530 struct llapi_ladvise_hdr *k_ladvise_hdr;
4531 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4534 int alloc_size = sizeof(*k_ladvise_hdr);
4537 u_ladvise_hdr = uarg;
4538 OBD_ALLOC_PTR(k_ladvise_hdr);
4539 if (k_ladvise_hdr == NULL)
4542 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4543 GOTO(out_ladvise, rc = -EFAULT);
4545 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4546 k_ladvise_hdr->lah_count < 1)
4547 GOTO(out_ladvise, rc = -EINVAL);
4549 num_advise = k_ladvise_hdr->lah_count;
4550 if (num_advise >= LAH_COUNT_MAX)
4551 GOTO(out_ladvise, rc = -EFBIG);
4553 OBD_FREE_PTR(k_ladvise_hdr);
4554 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4555 lah_advise[num_advise]);
4556 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4557 if (k_ladvise_hdr == NULL)
4561 * TODO: submit multiple advices to one server in a single RPC
4563 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4564 GOTO(out_ladvise, rc = -EFAULT);
4566 for (i = 0; i < num_advise; i++) {
4567 struct llapi_lu_ladvise *k_ladvise =
4568 &k_ladvise_hdr->lah_advise[i];
4569 struct llapi_lu_ladvise __user *u_ladvise =
4570 &u_ladvise_hdr->lah_advise[i];
4572 rc = ll_ladvise_sanity(inode, k_ladvise);
4574 GOTO(out_ladvise, rc);
4576 switch (k_ladvise->lla_advice) {
4577 case LU_LADVISE_LOCKNOEXPAND:
4578 rc = ll_lock_noexpand(file,
4579 k_ladvise->lla_peradvice_flags);
4580 GOTO(out_ladvise, rc);
4581 case LU_LADVISE_LOCKAHEAD:
4583 rc = ll_file_lock_ahead(file, k_ladvise);
4586 GOTO(out_ladvise, rc);
4589 &u_ladvise->lla_lockahead_result))
4590 GOTO(out_ladvise, rc = -EFAULT);
4593 rc = ll_ladvise(inode, file,
4594 k_ladvise_hdr->lah_flags,
4597 GOTO(out_ladvise, rc);
4604 OBD_FREE(k_ladvise_hdr, alloc_size);
4607 case LL_IOC_FLR_SET_MIRROR: {
4608 /* mirror I/O must be direct to avoid polluting page cache
4610 if (!(file->f_flags & O_DIRECT))
4613 fd->fd_designated_mirror = arg;
4616 case LL_IOC_HEAT_GET: {
4617 struct lu_heat uheat;
4618 struct lu_heat *heat;
4621 if (copy_from_user(&uheat, uarg, sizeof(uheat)))
4624 if (uheat.lh_count > OBD_HEAT_COUNT)
4625 uheat.lh_count = OBD_HEAT_COUNT;
4627 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4628 OBD_ALLOC(heat, size);
4632 heat->lh_count = uheat.lh_count;
4633 ll_heat_get(inode, heat);
4634 rc = copy_to_user(uarg, heat, size);
4635 OBD_FREE(heat, size);
4636 RETURN(rc ? -EFAULT : 0);
4638 case LL_IOC_HEAT_SET: {
4641 if (copy_from_user(&flags, uarg, sizeof(flags)))
4644 rc = ll_heat_set(inode, flags);
4647 case LL_IOC_PCC_DETACH: {
4648 struct lu_pcc_detach *detach;
4650 OBD_ALLOC_PTR(detach);
4654 if (copy_from_user(detach, uarg, sizeof(*detach)))
4655 GOTO(out_detach_free, rc = -EFAULT);
4657 if (!S_ISREG(inode->i_mode))
4658 GOTO(out_detach_free, rc = -EINVAL);
4660 if (!inode_owner_or_capable(&nop_mnt_idmap, inode))
4661 GOTO(out_detach_free, rc = -EPERM);
4663 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4665 OBD_FREE_PTR(detach);
4668 case LL_IOC_PCC_STATE: {
4669 struct lu_pcc_state __user *ustate = uarg;
4670 struct lu_pcc_state *state;
4672 OBD_ALLOC_PTR(state);
4676 if (copy_from_user(state, ustate, sizeof(*state)))
4677 GOTO(out_state, rc = -EFAULT);
4679 rc = pcc_ioctl_state(file, inode, state);
4681 GOTO(out_state, rc);
4683 if (copy_to_user(ustate, state, sizeof(*state)))
4684 GOTO(out_state, rc = -EFAULT);
4687 OBD_FREE_PTR(state);
4691 rc = ll_iocontrol(inode, file, cmd, uarg);
4694 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL, uarg));
4698 static loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4700 struct inode *inode = file_inode(file);
4703 struct cl_lseek_io *lsio;
4710 env = cl_env_get(&refcheck);
4712 RETURN(PTR_ERR(env));
4714 io = vvp_env_thread_io(env);
4715 io->ci_obj = ll_i2info(inode)->lli_clob;
4716 ll_io_set_mirror(io, file);
4718 lsio = &io->u.ci_lseek;
4719 lsio->ls_start = offset;
4720 lsio->ls_whence = whence;
4721 lsio->ls_result = -ENXIO;
4724 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4726 struct vvp_io *vio = vvp_env_io(env);
4728 vio->vui_fd = file->private_data;
4729 rc = cl_io_loop(env, io);
4733 retval = rc ? : lsio->ls_result;
4734 cl_io_fini(env, io);
4735 } while (unlikely(io->ci_need_restart));
4737 cl_env_put(env, &refcheck);
4739 /* Without the key, SEEK_HOLE return value has to be
4740 * rounded up to next LUSTRE_ENCRYPTION_UNIT_SIZE.
4742 if (llcrypt_require_key(inode) == -ENOKEY && whence == SEEK_HOLE)
4743 retval = round_up(retval, LUSTRE_ENCRYPTION_UNIT_SIZE);
4748 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4750 struct inode *inode = file_inode(file);
4751 loff_t retval = offset, eof = 0;
4752 ktime_t kstart = ktime_get();
4756 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4757 PFID(ll_inode2fid(inode)), inode, retval, retval,
4760 if (origin == SEEK_END) {
4761 retval = ll_glimpse_size(inode);
4764 eof = i_size_read(inode);
4767 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4771 /* flush local cache first if any */
4772 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4775 retval = ll_lseek(file, offset, origin);
4778 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4780 retval = generic_file_llseek_size(file, offset, origin,
4781 ll_file_maxbytes(inode), eof);
4784 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4785 ktime_us_delta(ktime_get(), kstart));
4789 static int ll_flush(struct file *file, fl_owner_t id)
4791 struct inode *inode = file_inode(file);
4792 struct ll_inode_info *lli = ll_i2info(inode);
4793 struct ll_file_data *fd = file->private_data;
4796 LASSERT(!S_ISDIR(inode->i_mode));
4798 /* catch async errors that were recorded back when async writeback
4799 * failed for pages in this mapping. */
4800 rc = lli->lli_async_rc;
4801 lli->lli_async_rc = 0;
4802 if (lli->lli_clob != NULL) {
4803 err = lov_read_and_clear_async_rc(lli->lli_clob);
4808 /* The application has been told write failure already.
4809 * Do not report failure again. */
4810 if (fd->fd_write_failed)
4812 return rc ? -EIO : 0;
4816 * Called to make sure a portion of file has been written out.
4817 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4819 * Return how many pages have been written.
4821 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4822 enum cl_fsync_mode mode, int ignore_layout)
4826 struct cl_fsync_io *fio;
4831 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4832 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL &&
4833 mode != CL_FSYNC_RECLAIM)
4836 env = cl_env_get(&refcheck);
4838 RETURN(PTR_ERR(env));
4840 io = vvp_env_thread_io(env);
4841 io->ci_obj = ll_i2info(inode)->lli_clob;
4842 io->ci_ignore_layout = ignore_layout;
4844 /* initialize parameters for sync */
4845 fio = &io->u.ci_fsync;
4846 fio->fi_start = start;
4848 fio->fi_fid = ll_inode2fid(inode);
4849 fio->fi_mode = mode;
4850 fio->fi_nr_written = 0;
4852 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4853 result = cl_io_loop(env, io);
4855 result = io->ci_result;
4857 result = fio->fi_nr_written;
4858 cl_io_fini(env, io);
4859 cl_env_put(env, &refcheck);
4865 * When dentry is provided (the 'else' case), file_dentry() may be
4866 * null and dentry must be used directly rather than pulled from
4867 * file_dentry() as is done otherwise.
4870 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4872 struct dentry *dentry = file_dentry(file);
4873 struct inode *inode = dentry->d_inode;
4874 struct ll_inode_info *lli = ll_i2info(inode);
4875 struct ptlrpc_request *req;
4876 ktime_t kstart = ktime_get();
4882 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4883 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4885 /* fsync's caller has already called _fdata{sync,write}, we want
4886 * that IO to finish before calling the osc and mdc sync methods */
4887 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4889 /* catch async errors that were recorded back when async writeback
4890 * failed for pages in this mapping. */
4891 if (!S_ISDIR(inode->i_mode)) {
4892 err = lli->lli_async_rc;
4893 lli->lli_async_rc = 0;
4896 if (lli->lli_clob != NULL) {
4897 err = lov_read_and_clear_async_rc(lli->lli_clob);
4903 if (S_ISREG(inode->i_mode) && !lli->lli_synced_to_mds) {
4905 * only the first sync on MDS makes sense,
4906 * everything else is stored on OSTs
4908 err = md_fsync(ll_i2sbi(inode)->ll_md_exp,
4909 ll_inode2fid(inode), &req);
4913 lli->lli_synced_to_mds = true;
4914 ptlrpc_req_finished(req);
4918 if (S_ISREG(inode->i_mode)) {
4919 struct ll_file_data *fd = file->private_data;
4922 /* Sync metadata on MDT first, and then sync the cached data
4925 err = pcc_fsync(file, start, end, datasync, &cached);
4927 err = cl_sync_file_range(inode, start, end,
4929 if (rc == 0 && err < 0)
4932 fd->fd_write_failed = true;
4934 fd->fd_write_failed = false;
4938 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4939 ktime_us_delta(ktime_get(), kstart));
4944 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4946 struct inode *inode = file_inode(file);
4947 struct ll_sb_info *sbi = ll_i2sbi(inode);
4948 struct ldlm_enqueue_info einfo = {
4949 .ei_type = LDLM_FLOCK,
4950 .ei_cb_cp = ldlm_flock_completion_ast,
4951 .ei_cbdata = file_lock,
4953 struct md_op_data *op_data;
4954 struct lustre_handle lockh = { 0 };
4955 union ldlm_policy_data flock = { { 0 } };
4956 int fl_type = file_lock->fl_type;
4957 ktime_t kstart = ktime_get();
4963 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4964 PFID(ll_inode2fid(inode)), file_lock);
4966 if (file_lock->fl_flags & FL_FLOCK) {
4967 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4968 /* flocks are whole-file locks */
4969 flock.l_flock.end = OFFSET_MAX;
4970 /* For flocks owner is determined by the local file desctiptor*/
4971 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4972 } else if (file_lock->fl_flags & FL_POSIX) {
4973 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4974 flock.l_flock.start = file_lock->fl_start;
4975 flock.l_flock.end = file_lock->fl_end;
4979 flock.l_flock.pid = file_lock->fl_pid;
4981 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4982 /* Somewhat ugly workaround for svc lockd.
4983 * lockd installs custom fl_lmops->lm_compare_owner that checks
4984 * for the fl_owner to be the same (which it always is on local node
4985 * I guess between lockd processes) and then compares pid.
4986 * As such we assign pid to the owner field to make it all work,
4987 * conflict with normal locks is unlikely since pid space and
4988 * pointer space for current->files are not intersecting */
4989 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4990 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4995 einfo.ei_mode = LCK_PR;
4998 /* An unlock request may or may not have any relation to
4999 * existing locks so we may not be able to pass a lock handle
5000 * via a normal ldlm_lock_cancel() request. The request may even
5001 * unlock a byte range in the middle of an existing lock. In
5002 * order to process an unlock request we need all of the same
5003 * information that is given with a normal read or write record
5004 * lock request. To avoid creating another ldlm unlock (cancel)
5005 * message we'll treat a LCK_NL flock request as an unlock. */
5006 einfo.ei_mode = LCK_NL;
5009 einfo.ei_mode = LCK_PW;
5013 CERROR("%s: fcntl from '%s' unknown lock type=%d: rc = %d\n",
5014 sbi->ll_fsname, current->comm, fl_type, rc);
5029 flags = LDLM_FL_BLOCK_NOWAIT;
5035 flags = LDLM_FL_TEST_LOCK;
5039 CERROR("%s: fcntl from '%s' unknown lock command=%d: rc = %d\n",
5040 sbi->ll_fsname, current->comm, cmd, rc);
5044 /* Save the old mode so that if the mode in the lock changes we
5045 * can decrement the appropriate reader or writer refcount. */
5046 file_lock->fl_type = einfo.ei_mode;
5048 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
5049 LUSTRE_OPC_ANY, NULL);
5050 if (IS_ERR(op_data))
5051 RETURN(PTR_ERR(op_data));
5053 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
5054 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
5055 flock.l_flock.pid, flags, einfo.ei_mode,
5056 flock.l_flock.start, flock.l_flock.end);
5058 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
5061 /* Restore the file lock type if not TEST lock. */
5062 if (!(flags & LDLM_FL_TEST_LOCK))
5063 file_lock->fl_type = fl_type;
5065 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
5066 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
5067 !(flags & LDLM_FL_TEST_LOCK))
5068 rc2 = locks_lock_file_wait(file, file_lock);
5070 if ((file_lock->fl_flags & FL_FLOCK) &&
5071 (rc == 0 || file_lock->fl_type == F_UNLCK))
5072 rc2 = flock_lock_file_wait(file, file_lock);
5073 if ((file_lock->fl_flags & FL_POSIX) &&
5074 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
5075 !(flags & LDLM_FL_TEST_LOCK))
5076 rc2 = posix_lock_file_wait(file, file_lock);
5077 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
5079 if (rc2 && file_lock->fl_type != F_UNLCK) {
5080 einfo.ei_mode = LCK_NL;
5081 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
5086 ll_finish_md_op_data(op_data);
5089 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
5090 ktime_us_delta(ktime_get(), kstart));
5094 int ll_get_fid_by_name(struct inode *parent, const char *name,
5095 int namelen, struct lu_fid *fid,
5096 struct inode **inode)
5098 struct md_op_data *op_data = NULL;
5099 struct mdt_body *body;
5100 struct ptlrpc_request *req;
5104 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
5105 LUSTRE_OPC_ANY, NULL);
5106 if (IS_ERR(op_data))
5107 RETURN(PTR_ERR(op_data));
5109 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
5110 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
5111 ll_finish_md_op_data(op_data);
5115 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
5117 GOTO(out_req, rc = -EFAULT);
5119 *fid = body->mbo_fid1;
5122 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
5124 ptlrpc_req_finished(req);
5128 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
5129 const char *name, __u32 flags)
5131 struct dentry *dchild = NULL;
5132 struct inode *child_inode = NULL;
5133 struct md_op_data *op_data;
5134 struct ptlrpc_request *request = NULL;
5135 struct obd_client_handle *och = NULL;
5137 struct mdt_body *body;
5138 __u64 data_version = 0;
5139 size_t namelen = strlen(name);
5140 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
5144 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
5145 PFID(ll_inode2fid(parent)), name,
5146 lum->lum_stripe_offset, lum->lum_stripe_count);
5148 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
5149 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
5150 lustre_swab_lmv_user_md(lum);
5152 /* Get child FID first */
5153 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
5156 dchild = d_lookup(file_dentry(file), &qstr);
5158 if (dchild->d_inode)
5159 child_inode = igrab(dchild->d_inode);
5164 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
5173 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
5174 OBD_CONNECT2_DIR_MIGRATE)) {
5175 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
5176 ll_dir_striped(child_inode)) {
5177 CERROR("%s: MDT doesn't support stripe directory "
5178 "migration!\n", ll_i2sbi(parent)->ll_fsname);
5179 GOTO(out_iput, rc = -EOPNOTSUPP);
5184 * lfs migrate command needs to be blocked on the client
5185 * by checking the migrate FID against the FID of the
5188 if (is_root_inode(child_inode))
5189 GOTO(out_iput, rc = -EINVAL);
5191 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
5192 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
5193 if (IS_ERR(op_data))
5194 GOTO(out_iput, rc = PTR_ERR(op_data));
5196 ll_inode_lock(child_inode);
5197 op_data->op_fid3 = *ll_inode2fid(child_inode);
5198 if (!fid_is_sane(&op_data->op_fid3)) {
5199 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
5200 ll_i2sbi(parent)->ll_fsname, name,
5201 PFID(&op_data->op_fid3));
5202 GOTO(out_unlock, rc = -EINVAL);
5205 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
5206 op_data->op_data = lum;
5207 op_data->op_data_size = lumlen;
5209 /* migrate dirent only for subdirs if MDS_MIGRATE_NSONLY set */
5210 if (S_ISDIR(child_inode->i_mode) && (flags & MDS_MIGRATE_NSONLY) &&
5211 lmv_dir_layout_changing(op_data->op_lso1))
5212 op_data->op_bias |= MDS_MIGRATE_NSONLY;
5215 if (S_ISREG(child_inode->i_mode)) {
5216 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
5220 GOTO(out_unlock, rc);
5223 rc = ll_data_version(child_inode, &data_version,
5226 GOTO(out_close, rc);
5228 op_data->op_open_handle = och->och_open_handle;
5229 op_data->op_data_version = data_version;
5230 op_data->op_lease_handle = och->och_lease_handle;
5231 op_data->op_bias |= MDS_CLOSE_MIGRATE;
5233 spin_lock(&och->och_mod->mod_open_req->rq_lock);
5234 och->och_mod->mod_open_req->rq_replay = 0;
5235 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
5238 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
5239 op_data->op_name, op_data->op_namelen,
5240 op_data->op_name, op_data->op_namelen, &request);
5242 LASSERT(request != NULL);
5243 ll_update_times(request, parent);
5246 if (rc == 0 || rc == -EAGAIN) {
5247 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
5248 LASSERT(body != NULL);
5250 /* If the server does release layout lock, then we cleanup
5251 * the client och here, otherwise release it in out_close: */
5252 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
5253 obd_mod_put(och->och_mod);
5254 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
5256 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
5262 if (request != NULL) {
5263 ptlrpc_req_finished(request);
5267 /* Try again if the lease has cancelled. */
5268 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
5273 ll_lease_close(och, child_inode, NULL);
5275 clear_nlink(child_inode);
5277 ll_inode_unlock(child_inode);
5278 ll_finish_md_op_data(op_data);
5285 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
5287 struct ll_file_data *fd = file->private_data;
5291 * In order to avoid flood of warning messages, only print one message
5292 * for one file. And the entire message rate on the client is limited
5293 * by CDEBUG_LIMIT too.
5295 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
5296 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
5297 CDEBUG_LIMIT(D_CONSOLE,
5298 "flock disabled, mount with '-o [local]flock' to enable\r\n");
5304 * test if some locks matching bits and l_req_mode are acquired
5305 * - bits can be in different locks
5306 * - if found clear the common lock bits in *bits
5307 * - the bits not found, are kept in *bits
5309 * \param bits [IN] searched lock bits [IN]
5310 * \param l_req_mode [IN] searched lock mode
5311 * \retval boolean, true iff all bits are found
5313 int ll_have_md_lock(struct obd_export *exp, struct inode *inode, __u64 *bits,
5314 enum ldlm_mode l_req_mode)
5316 struct lustre_handle lockh;
5317 union ldlm_policy_data policy;
5318 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
5319 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5328 fid = &ll_i2info(inode)->lli_fid;
5329 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5330 ldlm_lockname[mode]);
5332 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5333 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5334 policy.l_inodebits.bits = *bits & BIT(i);
5335 if (policy.l_inodebits.bits == 0)
5338 if (md_lock_match(exp, flags, fid, LDLM_IBITS, &policy, mode,
5340 struct ldlm_lock *lock;
5342 lock = ldlm_handle2lock(&lockh);
5345 ~(lock->l_policy_data.l_inodebits.bits);
5346 LDLM_LOCK_PUT(lock);
5348 *bits &= ~policy.l_inodebits.bits;
5355 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5356 struct lustre_handle *lockh, __u64 flags,
5357 enum ldlm_mode mode)
5359 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5364 fid = &ll_i2info(inode)->lli_fid;
5365 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5367 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5368 fid, LDLM_IBITS, &policy, mode, lockh);
5373 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5375 /* Already unlinked. Just update nlink and return success */
5376 if (rc == -ENOENT) {
5378 /* If it is striped directory, and there is bad stripe
5379 * Let's revalidate the dentry again, instead of returning
5381 if (ll_dir_striped(inode))
5384 /* This path cannot be hit for regular files unless in
5385 * case of obscure races, so no need to to validate
5387 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5389 } else if (rc != 0) {
5390 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5391 "%s: revalidate FID "DFID" error: rc = %d\n",
5392 ll_i2sbi(inode)->ll_fsname,
5393 PFID(ll_inode2fid(inode)), rc);
5399 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5401 struct inode *parent;
5402 struct inode *inode = dentry->d_inode;
5403 struct obd_export *exp = ll_i2mdexp(inode);
5404 struct lookup_intent oit = {
5407 struct ptlrpc_request *req = NULL;
5408 struct md_op_data *op_data;
5409 const char *name = NULL;
5414 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5415 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5417 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5418 parent = dentry->d_parent->d_inode;
5419 name = dentry->d_name.name;
5420 namelen = dentry->d_name.len;
5425 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5426 LUSTRE_OPC_ANY, NULL);
5427 if (IS_ERR(op_data))
5428 RETURN(PTR_ERR(op_data));
5430 /* Call getattr by fid */
5431 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5432 op_data->op_flags = MF_GETATTR_BY_FID;
5433 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5434 ll_finish_md_op_data(op_data);
5436 rc = ll_inode_revalidate_fini(inode, rc);
5440 rc = ll_revalidate_it_finish(req, &oit, dentry);
5442 ll_intent_release(&oit);
5446 /* Unlinked? Unhash dentry, so it is not picked up later by
5447 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5448 * here to preserve get_cwd functionality on 2.6.
5450 if (!dentry->d_inode->i_nlink)
5451 d_lustre_invalidate(dentry);
5453 ll_lookup_finish_locks(&oit, dentry);
5455 ptlrpc_req_finished(req);
5460 static int ll_merge_md_attr(struct inode *inode)
5462 struct ll_inode_info *lli = ll_i2info(inode);
5463 struct lmv_stripe_object *lsm_obj;
5464 struct cl_attr attr = { 0 };
5467 if (!ll_dir_striped(inode))
5470 down_read(&lli->lli_lsm_sem);
5471 LASSERT(lli->lli_lsm_obj != NULL);
5473 lsm_obj = lmv_stripe_object_get(lli->lli_lsm_obj);
5474 up_read(&lli->lli_lsm_sem);
5476 rc = md_merge_attr(ll_i2mdexp(inode), lsm_obj,
5477 &attr, ll_md_blocking_ast);
5478 lmv_stripe_object_put(&lsm_obj);
5482 spin_lock(&inode->i_lock);
5483 set_nlink(inode, attr.cat_nlink);
5484 spin_unlock(&inode->i_lock);
5486 inode->i_blocks = attr.cat_blocks;
5487 i_size_write(inode, attr.cat_size);
5489 ll_i2info(inode)->lli_atime = attr.cat_atime;
5490 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5491 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5496 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5497 unsigned int flags, bool foreign)
5499 struct inode *inode = de->d_inode;
5500 struct ll_sb_info *sbi = ll_i2sbi(inode);
5501 struct ll_inode_info *lli = ll_i2info(inode);
5502 struct inode *dir = de->d_parent->d_inode;
5503 bool need_glimpse = true;
5504 ktime_t kstart = ktime_get();
5507 /* The OST object(s) determine the file size, blocks and mtime. */
5508 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5509 request_mask & STATX_MTIME))
5510 need_glimpse = false;
5512 ll_statahead_enter(dir, de);
5513 if (dentry_may_statahead(dir, de))
5514 ll_start_statahead(dir, de, need_glimpse &&
5515 !(flags & AT_STATX_DONT_SYNC));
5517 if (flags & AT_STATX_DONT_SYNC)
5518 GOTO(fill_attr, rc = 0);
5520 rc = ll_inode_revalidate(de, IT_GETATTR);
5524 /* foreign file/dir are always of zero length, so don't
5525 * need to validate size.
5527 if (S_ISREG(inode->i_mode) && !foreign) {
5531 GOTO(fill_attr, rc);
5533 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5534 if (cached && rc < 0)
5538 GOTO(fill_attr, rc);
5541 * If the returned attr is masked with OBD_MD_FLSIZE &
5542 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5543 * or blocks obtained from MDT is strictly correct, and the file
5544 * is usually not being modified by clients, and the [a|m|c]time
5545 * got from MDT is also strictly correct.
5546 * Under this circumstance, it does not need to send glimpse
5547 * RPCs to OSTs for file attributes such as the size and blocks.
5549 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5550 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5551 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5552 inode->i_mtime.tv_sec = lli->lli_mtime;
5553 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5554 inode->i_atime.tv_sec = lli->lli_atime;
5555 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5556 inode->i_ctime.tv_sec = lli->lli_ctime;
5557 GOTO(fill_attr, rc);
5560 /* In case of restore, the MDT has the right size and has
5561 * already send it back without granting the layout lock,
5562 * inode is up-to-date so glimpse is useless.
5563 * Also to glimpse we need the layout, in case of a running
5564 * restore the MDT holds the layout lock so the glimpse will
5565 * block up to the end of restore (getattr will block)
5567 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5568 rc = ll_glimpse_size(inode);
5573 /* If object isn't regular a file then don't validate size. */
5574 /* foreign dir is not striped dir */
5576 rc = ll_merge_md_attr(inode);
5581 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5582 inode->i_atime.tv_sec = lli->lli_atime;
5583 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5584 inode->i_mtime.tv_sec = lli->lli_mtime;
5585 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5586 inode->i_ctime.tv_sec = lli->lli_ctime;
5590 CFS_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5592 if (ll_need_32bit_api(sbi)) {
5593 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5594 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5595 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5597 stat->ino = inode->i_ino;
5598 stat->dev = inode->i_sb->s_dev;
5599 stat->rdev = inode->i_rdev;
5602 /* foreign symlink to be exposed as a real symlink */
5604 stat->mode = inode->i_mode;
5606 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5608 stat->uid = inode->i_uid;
5609 stat->gid = inode->i_gid;
5610 stat->atime = inode->i_atime;
5611 stat->mtime = inode->i_mtime;
5612 stat->ctime = inode->i_ctime;
5613 /* stat->blksize is used to tell about preferred IO size */
5614 if (sbi->ll_stat_blksize)
5615 stat->blksize = sbi->ll_stat_blksize;
5616 else if (S_ISREG(inode->i_mode))
5617 stat->blksize = min(PTLRPC_MAX_BRW_SIZE,
5618 1U << LL_MAX_BLKSIZE_BITS);
5619 else if (S_ISDIR(inode->i_mode))
5620 stat->blksize = min(MD_MAX_BRW_SIZE,
5621 1U << LL_MAX_BLKSIZE_BITS);
5623 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5625 stat->nlink = inode->i_nlink;
5626 stat->size = i_size_read(inode);
5627 stat->blocks = inode->i_blocks;
5629 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5630 if (flags & AT_STATX_DONT_SYNC) {
5631 if (stat->size == 0 &&
5632 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5633 stat->size = lli->lli_lazysize;
5634 if (stat->blocks == 0 &&
5635 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5636 stat->blocks = lli->lli_lazyblocks;
5639 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5640 stat->result_mask |= STATX_BTIME;
5641 stat->btime.tv_sec = lli->lli_btime;
5644 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5645 #ifdef HAVE_LUSTRE_CRYPTO
5646 stat->attributes_mask |= STATX_ATTR_ENCRYPTED;
5648 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5649 /* if Lustre specific LUSTRE_ENCRYPT_FL flag is set, also set
5650 * ext4 equivalent to please statx
5652 if (stat->attributes & LUSTRE_ENCRYPT_FL)
5653 stat->attributes |= STATX_ATTR_ENCRYPTED;
5654 stat->result_mask &= request_mask;
5657 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5658 ktime_us_delta(ktime_get(), kstart));
5663 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5664 int ll_getattr(struct mnt_idmap *map, const struct path *path,
5665 struct kstat *stat, u32 request_mask, unsigned int flags)
5667 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5671 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5673 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5674 AT_STATX_SYNC_AS_STAT, false);
5678 static int cl_falloc(struct file *file, struct inode *inode, int mode,
5679 loff_t offset, loff_t len)
5681 loff_t size = i_size_read(inode);
5689 env = cl_env_get(&refcheck);
5691 RETURN(PTR_ERR(env));
5693 io = vvp_env_thread_io(env);
5694 io->ci_obj = ll_i2info(inode)->lli_clob;
5695 ll_io_set_mirror(io, file);
5697 io->ci_verify_layout = 1;
5698 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5699 io->u.ci_setattr.sa_falloc_mode = mode;
5700 io->u.ci_setattr.sa_falloc_offset = offset;
5701 io->u.ci_setattr.sa_falloc_end = offset + len;
5702 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5704 CDEBUG(D_INODE, "UID %u GID %u PRJID %u\n",
5705 from_kuid(&init_user_ns, inode->i_uid),
5706 from_kgid(&init_user_ns, inode->i_gid),
5707 ll_i2info(inode)->lli_projid);
5709 io->u.ci_setattr.sa_falloc_uid = from_kuid(&init_user_ns, inode->i_uid);
5710 io->u.ci_setattr.sa_falloc_gid = from_kgid(&init_user_ns, inode->i_gid);
5711 io->u.ci_setattr.sa_falloc_projid = ll_i2info(inode)->lli_projid;
5713 if (io->u.ci_setattr.sa_falloc_end > size) {
5714 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5716 /* Check new size against VFS/VM file size limit and rlimit */
5717 rc = inode_newsize_ok(inode, newsize);
5720 if (newsize > ll_file_maxbytes(inode)) {
5721 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5722 (unsigned long long)newsize,
5723 ll_file_maxbytes(inode));
5730 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5732 rc = cl_io_loop(env, io);
5735 cl_io_fini(env, io);
5736 } while (unlikely(io->ci_need_restart));
5739 cl_env_put(env, &refcheck);
5743 static long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5745 struct inode *inode = file_inode(filp);
5748 if (offset < 0 || len <= 0)
5751 * Encrypted inodes can't handle collapse range or zero range or insert
5752 * range since we would need to re-encrypt blocks with a different IV or
5753 * XTS tweak (which are based on the logical block number).
5754 * Similar to what ext4 does.
5756 if (IS_ENCRYPTED(inode) &&
5757 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5758 FALLOC_FL_ZERO_RANGE)))
5759 RETURN(-EOPNOTSUPP);
5762 * mode == 0 (which is standard prealloc) and PUNCH is supported
5763 * Rest of mode options are not supported yet.
5765 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5766 RETURN(-EOPNOTSUPP);
5768 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5770 rc = cl_falloc(filp, inode, mode, offset, len);
5772 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5773 * used by Lustre in several places. Retuning it here would
5774 * confuse applications that explicity test for EOPNOTSUPP
5775 * (95) and fall back to ftruncate().
5777 if (rc == -ENOTSUPP)
5783 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5784 __u64 start, __u64 len)
5788 struct fiemap *fiemap;
5789 unsigned int extent_count = fieinfo->fi_extents_max;
5791 num_bytes = sizeof(*fiemap) + (extent_count *
5792 sizeof(struct fiemap_extent));
5793 OBD_ALLOC_LARGE(fiemap, num_bytes);
5798 fiemap->fm_flags = fieinfo->fi_flags;
5799 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5800 fiemap->fm_start = start;
5801 fiemap->fm_length = len;
5802 if (extent_count > 0 &&
5803 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5804 sizeof(struct fiemap_extent)) != 0)
5805 GOTO(out, rc = -EFAULT);
5807 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5809 if (IS_ENCRYPTED(inode) && extent_count > 0) {
5812 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5813 fiemap->fm_extents[i].fe_flags |=
5814 FIEMAP_EXTENT_DATA_ENCRYPTED |
5815 FIEMAP_EXTENT_ENCODED;
5818 fieinfo->fi_flags = fiemap->fm_flags;
5819 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5820 if (extent_count > 0 &&
5821 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5822 fiemap->fm_mapped_extents *
5823 sizeof(struct fiemap_extent)) != 0)
5824 GOTO(out, rc = -EFAULT);
5826 OBD_FREE_LARGE(fiemap, num_bytes);
5830 int ll_inode_permission(struct mnt_idmap *idmap, struct inode *inode, int mask)
5833 struct ll_sb_info *sbi;
5834 struct root_squash_info *squash;
5835 struct cred *cred = NULL;
5836 const struct cred *old_cred = NULL;
5837 bool squash_id = false;
5838 ktime_t kstart = ktime_get();
5842 if (mask & MAY_NOT_BLOCK)
5846 * as root inode are NOT getting validated in lookup operation,
5847 * need to revalidate PERM before permission check.
5849 if (is_root_inode(inode)) {
5850 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_GETATTR);
5855 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5856 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5858 /* squash fsuid/fsgid if needed */
5859 sbi = ll_i2sbi(inode);
5860 squash = &sbi->ll_squash;
5861 if (unlikely(squash->rsi_uid != 0 &&
5862 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5863 !test_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags))) {
5867 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5868 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5869 squash->rsi_uid, squash->rsi_gid);
5871 /* update current process's credentials
5872 * and FS capability */
5873 cred = prepare_creds();
5877 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5878 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5879 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5880 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5882 old_cred = override_creds(cred);
5885 rc = generic_permission(idmap, inode, mask);
5886 /* restore current process's credentials and FS capability */
5888 revert_creds(old_cred);
5893 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5894 ktime_us_delta(ktime_get(), kstart));
5899 /* -o localflock - only provides locally consistent flock locks */
5900 static const struct file_operations ll_file_operations = {
5901 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5902 # ifdef HAVE_SYNC_READ_WRITE
5903 .read = new_sync_read,
5904 .write = new_sync_write,
5906 .read_iter = ll_file_read_iter,
5907 .write_iter = ll_file_write_iter,
5908 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5909 .read = ll_file_read,
5910 .aio_read = ll_file_aio_read,
5911 .write = ll_file_write,
5912 .aio_write = ll_file_aio_write,
5913 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5914 .unlocked_ioctl = ll_file_ioctl,
5915 .open = ll_file_open,
5916 .release = ll_file_release,
5917 .mmap = ll_file_mmap,
5918 .llseek = ll_file_seek,
5919 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5920 .splice_read = generic_file_splice_read,
5922 .splice_read = pcc_file_splice_read,
5924 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5925 .splice_write = iter_file_splice_write,
5929 .fallocate = ll_fallocate,
5932 static const struct file_operations ll_file_operations_flock = {
5933 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5934 # ifdef HAVE_SYNC_READ_WRITE
5935 .read = new_sync_read,
5936 .write = new_sync_write,
5937 # endif /* HAVE_SYNC_READ_WRITE */
5938 .read_iter = ll_file_read_iter,
5939 .write_iter = ll_file_write_iter,
5940 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5941 .read = ll_file_read,
5942 .aio_read = ll_file_aio_read,
5943 .write = ll_file_write,
5944 .aio_write = ll_file_aio_write,
5945 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5946 .unlocked_ioctl = ll_file_ioctl,
5947 .open = ll_file_open,
5948 .release = ll_file_release,
5949 .mmap = ll_file_mmap,
5950 .llseek = ll_file_seek,
5951 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5952 .splice_read = generic_file_splice_read,
5954 .splice_read = pcc_file_splice_read,
5956 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5957 .splice_write = iter_file_splice_write,
5961 .flock = ll_file_flock,
5962 .lock = ll_file_flock,
5963 .fallocate = ll_fallocate,
5966 /* These are for -o noflock - to return ENOSYS on flock calls */
5967 static const struct file_operations ll_file_operations_noflock = {
5968 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5969 # ifdef HAVE_SYNC_READ_WRITE
5970 .read = new_sync_read,
5971 .write = new_sync_write,
5972 # endif /* HAVE_SYNC_READ_WRITE */
5973 .read_iter = ll_file_read_iter,
5974 .write_iter = ll_file_write_iter,
5975 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5976 .read = ll_file_read,
5977 .aio_read = ll_file_aio_read,
5978 .write = ll_file_write,
5979 .aio_write = ll_file_aio_write,
5980 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5981 .unlocked_ioctl = ll_file_ioctl,
5982 .open = ll_file_open,
5983 .release = ll_file_release,
5984 .mmap = ll_file_mmap,
5985 .llseek = ll_file_seek,
5986 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5987 .splice_read = generic_file_splice_read,
5989 .splice_read = pcc_file_splice_read,
5991 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5992 .splice_write = iter_file_splice_write,
5996 .flock = ll_file_noflock,
5997 .lock = ll_file_noflock,
5998 .fallocate = ll_fallocate,
6001 const struct inode_operations ll_file_inode_operations = {
6002 .setattr = ll_setattr,
6003 .getattr = ll_getattr,
6004 .permission = ll_inode_permission,
6005 #ifdef HAVE_IOP_XATTR
6006 .setxattr = ll_setxattr,
6007 .getxattr = ll_getxattr,
6008 .removexattr = ll_removexattr,
6010 .listxattr = ll_listxattr,
6011 .fiemap = ll_fiemap,
6012 .get_acl = ll_get_acl,
6013 #ifdef HAVE_IOP_SET_ACL
6014 .set_acl = ll_set_acl,
6016 #ifdef HAVE_FILEATTR_GET
6017 .fileattr_get = ll_fileattr_get,
6018 .fileattr_set = ll_fileattr_set,
6022 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
6024 const struct file_operations *fops = &ll_file_operations_noflock;
6026 if (test_bit(LL_SBI_FLOCK, sbi->ll_flags))
6027 fops = &ll_file_operations_flock;
6028 else if (test_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags))
6029 fops = &ll_file_operations;
6034 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
6036 struct ll_inode_info *lli = ll_i2info(inode);
6037 struct cl_object *obj = lli->lli_clob;
6046 env = cl_env_get(&refcheck);
6048 RETURN(PTR_ERR(env));
6050 rc = cl_conf_set(env, lli->lli_clob, conf);
6054 if (conf->coc_opc == OBJECT_CONF_SET) {
6055 struct ldlm_lock *lock = conf->coc_lock;
6056 struct cl_layout cl = {
6060 LASSERT(lock != NULL);
6061 LASSERT(ldlm_has_layout(lock));
6063 /* it can only be allowed to match after layout is
6064 * applied to inode otherwise false layout would be
6065 * seen. Applying layout shoud happen before dropping
6066 * the intent lock. */
6067 ldlm_lock_allow_match(lock);
6069 rc = cl_object_layout_get(env, obj, &cl);
6074 DFID": layout version change: %u -> %u\n",
6075 PFID(&lli->lli_fid), ll_layout_version_get(lli),
6077 ll_layout_version_set(lli, cl.cl_layout_gen);
6081 cl_env_put(env, &refcheck);
6083 RETURN(rc < 0 ? rc : 0);
6086 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
6087 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
6090 struct ll_sb_info *sbi = ll_i2sbi(inode);
6091 struct ptlrpc_request *req;
6098 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
6099 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
6100 lock->l_lvb_data, lock->l_lvb_len);
6102 if (lock->l_lvb_data != NULL)
6105 /* if layout lock was granted right away, the layout is returned
6106 * within DLM_LVB of dlm reply; otherwise if the lock was ever
6107 * blocked and then granted via completion ast, we have to fetch
6108 * layout here. Please note that we can't use the LVB buffer in
6109 * completion AST because it doesn't have a large enough buffer */
6110 rc = ll_get_default_mdsize(sbi, &lmmsize);
6114 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
6115 XATTR_NAME_LOV, lmmsize, &req);
6118 GOTO(out, rc = 0); /* empty layout */
6125 if (lmmsize == 0) /* empty layout */
6128 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
6130 GOTO(out, rc = -EFAULT);
6132 OBD_ALLOC_LARGE(lvbdata, lmmsize);
6133 if (lvbdata == NULL)
6134 GOTO(out, rc = -ENOMEM);
6136 memcpy(lvbdata, lmm, lmmsize);
6137 lock_res_and_lock(lock);
6138 if (unlikely(lock->l_lvb_data == NULL)) {
6139 lock->l_lvb_type = LVB_T_LAYOUT;
6140 lock->l_lvb_data = lvbdata;
6141 lock->l_lvb_len = lmmsize;
6144 unlock_res_and_lock(lock);
6147 OBD_FREE_LARGE(lvbdata, lmmsize);
6152 ptlrpc_req_finished(req);
6157 * Apply the layout to the inode. Layout lock is held and will be released
6160 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
6161 struct inode *inode, bool try)
6163 struct ll_inode_info *lli = ll_i2info(inode);
6164 struct ll_sb_info *sbi = ll_i2sbi(inode);
6165 struct ldlm_lock *lock;
6166 struct cl_object_conf conf;
6169 bool wait_layout = false;
6172 LASSERT(lustre_handle_is_used(lockh));
6174 lock = ldlm_handle2lock(lockh);
6175 LASSERT(lock != NULL);
6177 if (!ldlm_has_layout(lock))
6178 GOTO(out, rc = -EAGAIN);
6180 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
6181 PFID(&lli->lli_fid), inode);
6183 /* in case this is a caching lock and reinstate with new inode */
6184 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
6186 lock_res_and_lock(lock);
6187 lvb_ready = ldlm_is_lvb_ready(lock);
6188 unlock_res_and_lock(lock);
6190 /* checking lvb_ready is racy but this is okay. The worst case is
6191 * that multi processes may configure the file on the same time. */
6195 rc = ll_layout_fetch(inode, lock);
6199 /* for layout lock, lmm is stored in lock's lvb.
6200 * lvb_data is immutable if the lock is held so it's safe to access it
6203 * set layout to file. Unlikely this will fail as old layout was
6204 * surely eliminated */
6205 memset(&conf, 0, sizeof conf);
6206 conf.coc_opc = OBJECT_CONF_SET;
6207 conf.coc_inode = inode;
6208 conf.coc_lock = lock;
6210 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
6211 conf.u.coc_layout.lb_len = lock->l_lvb_len;
6212 rc = ll_layout_conf(inode, &conf);
6214 /* refresh layout failed, need to wait */
6215 wait_layout = rc == -EBUSY;
6218 LDLM_LOCK_PUT(lock);
6219 ldlm_lock_decref(lockh, mode);
6221 /* wait for IO to complete if it's still being used. */
6223 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
6224 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6226 memset(&conf, 0, sizeof conf);
6227 conf.coc_opc = OBJECT_CONF_WAIT;
6228 conf.coc_inode = inode;
6229 rc = ll_layout_conf(inode, &conf);
6233 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
6234 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
6237 if (rc == -ERESTARTSYS) {
6240 struct cl_object * obj = lli->lli_clob;
6242 env = cl_env_get(&refcheck);
6244 RETURN(PTR_ERR(env));
6246 CDEBUG(D_INODE, "prune without lock "DFID"\n",
6247 PFID(lu_object_fid(&obj->co_lu)));
6249 trunc_sem_down_write(&lli->lli_trunc_sem);
6250 cl_object_prune(env, obj);
6251 trunc_sem_up_write(&lli->lli_trunc_sem);
6252 cl_env_put(env, &refcheck);
6261 * Issue layout intent RPC to MDS.
6262 * \param inode [in] file inode
6263 * \param intent [in] layout intent
6265 * \retval 0 on success
6266 * \retval < 0 error code
6268 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
6270 struct ll_inode_info *lli = ll_i2info(inode);
6271 struct ll_sb_info *sbi = ll_i2sbi(inode);
6272 struct md_op_data *op_data;
6273 struct lookup_intent it;
6274 struct ptlrpc_request *req;
6278 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
6279 0, 0, LUSTRE_OPC_ANY, NULL);
6280 if (IS_ERR(op_data))
6281 RETURN(PTR_ERR(op_data));
6283 op_data->op_data = intent;
6284 op_data->op_data_size = sizeof(*intent);
6286 memset(&it, 0, sizeof(it));
6287 it.it_op = IT_LAYOUT;
6288 if (intent->lai_opc == LAYOUT_INTENT_WRITE ||
6289 intent->lai_opc == LAYOUT_INTENT_TRUNC)
6290 it.it_flags = FMODE_WRITE;
6292 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
6293 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
6295 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
6296 &ll_md_blocking_ast, 0);
6297 if (it.it_request != NULL)
6298 ptlrpc_req_finished(it.it_request);
6299 it.it_request = NULL;
6301 ll_finish_md_op_data(op_data);
6303 /* set lock data in case this is a new lock */
6305 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
6307 ll_intent_drop_lock(&it);
6313 * This function checks if there exists a LAYOUT lock on the client side,
6314 * or enqueues it if it doesn't have one in cache.
6316 * This function will not hold layout lock so it may be revoked any time after
6317 * this function returns. Any operations depend on layout should be redone
6320 * This function should be called before lov_io_init() to get an uptodate
6321 * layout version, the caller should save the version number and after IO
6322 * is finished, this function should be called again to verify that layout
6323 * is not changed during IO time.
6325 int ll_layout_refresh(struct inode *inode, __u32 *gen)
6327 struct ll_inode_info *lli = ll_i2info(inode);
6328 struct ll_sb_info *sbi = ll_i2sbi(inode);
6329 struct lustre_handle lockh;
6330 struct layout_intent intent = {
6331 .lai_opc = LAYOUT_INTENT_ACCESS,
6333 enum ldlm_mode mode;
6338 *gen = ll_layout_version_get(lli);
6339 if (!test_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags) ||
6340 *gen != CL_LAYOUT_GEN_NONE)
6344 LASSERT(fid_is_sane(ll_inode2fid(inode)));
6345 LASSERT(S_ISREG(inode->i_mode));
6348 /* mostly layout lock is caching on the local side, so try to
6349 * match it before grabbing layout lock mutex. */
6350 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
6351 LCK_CR | LCK_CW | LCK_PR |
6353 if (mode != 0) { /* hit cached lock */
6354 rc = ll_layout_lock_set(&lockh, mode, inode, try);
6361 /* take layout lock mutex to enqueue layout lock exclusively. */
6362 mutex_lock(&lli->lli_layout_mutex);
6363 rc = ll_layout_intent(inode, &intent);
6364 mutex_unlock(&lli->lli_layout_mutex);
6370 *gen = ll_layout_version_get(lli);
6376 * Issue layout intent RPC indicating where in a file an IO is about to write.
6378 * \param[in] inode file inode.
6379 * \param[in] ext write range with start offset of fille in bytes where
6380 * an IO is about to write, and exclusive end offset in
6383 * \retval 0 on success
6384 * \retval < 0 error code
6386 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
6387 struct lu_extent *ext)
6389 struct layout_intent intent = {
6391 .lai_extent.e_start = ext->e_start,
6392 .lai_extent.e_end = ext->e_end,
6397 rc = ll_layout_intent(inode, &intent);
6403 * This function send a restore request to the MDT
6405 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6407 struct hsm_user_request *hur;
6411 len = sizeof(struct hsm_user_request) +
6412 sizeof(struct hsm_user_item);
6413 OBD_ALLOC(hur, len);
6417 hur->hur_request.hr_action = HUA_RESTORE;
6418 hur->hur_request.hr_archive_id = 0;
6419 hur->hur_request.hr_flags = 0;
6420 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6421 sizeof(hur->hur_user_item[0].hui_fid));
6422 hur->hur_user_item[0].hui_extent.offset = offset;
6423 hur->hur_user_item[0].hui_extent.length = length;
6424 hur->hur_request.hr_itemcount = 1;
6425 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,