4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
33 * Author: Peter Braam <braam@clusterfs.com>
34 * Author: Phil Schwan <phil@clusterfs.com>
35 * Author: Andreas Dilger <adilger@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <lustre_dlm.h>
40 #include <linux/pagemap.h>
41 #include <linux/file.h>
42 #include <linux/sched.h>
43 #include <linux/user_namespace.h>
44 #include <linux/uidgid.h>
45 #include <linux/falloc.h>
46 #include <linux/ktime.h>
48 #include <uapi/linux/lustre/lustre_ioctl.h>
49 #include <lustre_swab.h>
51 #include "cl_object.h"
52 #include "llite_internal.h"
53 #include "vvp_internal.h"
56 struct inode *sp_inode;
61 __u64 pa_data_version;
67 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
69 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
72 static struct ll_file_data *ll_file_data_get(void)
74 struct ll_file_data *fd;
76 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
80 fd->fd_write_failed = false;
81 pcc_file_init(&fd->fd_pcc_file);
86 static void ll_file_data_put(struct ll_file_data *fd)
89 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
93 * Packs all the attributes into @op_data for the CLOSE rpc.
95 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
96 struct obd_client_handle *och)
100 ll_prep_md_op_data(op_data, inode, NULL, NULL,
101 0, 0, LUSTRE_OPC_ANY, NULL);
103 op_data->op_attr.ia_mode = inode->i_mode;
104 op_data->op_attr.ia_atime = inode->i_atime;
105 op_data->op_attr.ia_mtime = inode->i_mtime;
106 op_data->op_attr.ia_ctime = inode->i_ctime;
107 /* In case of encrypted file without the key, visible size was rounded
108 * up to next LUSTRE_ENCRYPTION_UNIT_SIZE, and clear text size was
109 * stored into lli_lazysize in ll_merge_attr(), so set proper file size
110 * now that we are closing.
112 if (llcrypt_require_key(inode) == -ENOKEY &&
113 ll_i2info(inode)->lli_attr_valid & OBD_MD_FLLAZYSIZE)
114 op_data->op_attr.ia_size = ll_i2info(inode)->lli_lazysize;
116 op_data->op_attr.ia_size = i_size_read(inode);
117 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
118 ATTR_MTIME | ATTR_MTIME_SET |
120 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
121 op_data->op_attr_blocks = inode->i_blocks;
122 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
123 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
124 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
125 op_data->op_open_handle = och->och_open_handle;
127 if (och->och_flags & FMODE_WRITE &&
128 test_and_clear_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags))
129 /* For HSM: if inode data has been modified, pack it so that
130 * MDT can set data dirty flag in the archive. */
131 op_data->op_bias |= MDS_DATA_MODIFIED;
137 * Perform a close, possibly with a bias.
138 * The meaning of "data" depends on the value of "bias".
140 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
141 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
144 static int ll_close_inode_openhandle(struct inode *inode,
145 struct obd_client_handle *och,
146 enum mds_op_bias bias, void *data)
148 struct obd_export *md_exp = ll_i2mdexp(inode);
149 const struct ll_inode_info *lli = ll_i2info(inode);
150 struct md_op_data *op_data;
151 struct ptlrpc_request *req = NULL;
155 if (class_exp2obd(md_exp) == NULL) {
156 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
157 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
161 OBD_ALLOC_PTR(op_data);
162 /* We leak openhandle and request here on error, but not much to be
163 * done in OOM case since app won't retry close on error either. */
165 GOTO(out, rc = -ENOMEM);
167 ll_prepare_close(inode, op_data, och);
169 case MDS_CLOSE_LAYOUT_MERGE:
170 /* merge blocks from the victim inode */
171 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
172 op_data->op_attr.ia_valid |= ATTR_SIZE;
173 op_data->op_xvalid |= OP_XVALID_BLOCKS;
175 case MDS_CLOSE_LAYOUT_SPLIT:
176 case MDS_CLOSE_LAYOUT_SWAP: {
177 struct split_param *sp = data;
179 LASSERT(data != NULL);
180 op_data->op_bias |= bias;
181 op_data->op_data_version = 0;
182 op_data->op_lease_handle = och->och_lease_handle;
183 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
184 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
185 op_data->op_mirror_id = sp->sp_mirror_id;
187 op_data->op_fid2 = *ll_inode2fid(data);
192 case MDS_CLOSE_RESYNC_DONE: {
193 struct ll_ioc_lease *ioc = data;
195 LASSERT(data != NULL);
196 op_data->op_attr_blocks +=
197 ioc->lil_count * op_data->op_attr_blocks;
198 op_data->op_attr.ia_valid |= ATTR_SIZE;
199 op_data->op_xvalid |= OP_XVALID_BLOCKS;
200 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
202 op_data->op_lease_handle = och->och_lease_handle;
203 op_data->op_data = &ioc->lil_ids[0];
204 op_data->op_data_size =
205 ioc->lil_count * sizeof(ioc->lil_ids[0]);
209 case MDS_PCC_ATTACH: {
210 struct pcc_param *param = data;
212 LASSERT(data != NULL);
213 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
214 op_data->op_archive_id = param->pa_archive_id;
215 op_data->op_data_version = param->pa_data_version;
216 op_data->op_lease_handle = och->och_lease_handle;
220 case MDS_HSM_RELEASE:
221 LASSERT(data != NULL);
222 op_data->op_bias |= MDS_HSM_RELEASE;
223 op_data->op_data_version = *(__u64 *)data;
224 op_data->op_lease_handle = och->och_lease_handle;
225 op_data->op_attr.ia_valid |= ATTR_SIZE;
226 op_data->op_xvalid |= OP_XVALID_BLOCKS;
230 LASSERT(data == NULL);
234 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
235 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
236 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
237 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
239 rc = md_close(md_exp, op_data, och->och_mod, &req);
240 if (rc != 0 && rc != -EINTR)
241 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
242 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
244 if (rc == 0 && op_data->op_bias & bias) {
245 struct mdt_body *body;
247 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
248 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
251 if (bias & MDS_PCC_ATTACH) {
252 struct pcc_param *param = data;
254 param->pa_layout_gen = body->mbo_layout_gen;
258 ll_finish_md_op_data(op_data);
262 md_clear_open_replay_data(md_exp, och);
263 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
266 ptlrpc_req_finished(req); /* This is close request */
270 int ll_md_real_close(struct inode *inode, fmode_t fmode)
272 struct ll_inode_info *lli = ll_i2info(inode);
273 struct obd_client_handle **och_p;
274 struct obd_client_handle *och;
279 if (fmode & FMODE_WRITE) {
280 och_p = &lli->lli_mds_write_och;
281 och_usecount = &lli->lli_open_fd_write_count;
282 } else if (fmode & FMODE_EXEC) {
283 och_p = &lli->lli_mds_exec_och;
284 och_usecount = &lli->lli_open_fd_exec_count;
286 LASSERT(fmode & FMODE_READ);
287 och_p = &lli->lli_mds_read_och;
288 och_usecount = &lli->lli_open_fd_read_count;
291 mutex_lock(&lli->lli_och_mutex);
292 if (*och_usecount > 0) {
293 /* There are still users of this handle, so skip
295 mutex_unlock(&lli->lli_och_mutex);
301 mutex_unlock(&lli->lli_och_mutex);
304 /* There might be a race and this handle may already
306 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
312 static int ll_md_close(struct inode *inode, struct file *file)
314 union ldlm_policy_data policy = {
315 .l_inodebits = { MDS_INODELOCK_OPEN },
317 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
318 struct ll_file_data *fd = file->private_data;
319 struct ll_inode_info *lli = ll_i2info(inode);
320 struct lustre_handle lockh;
321 enum ldlm_mode lockmode;
325 /* clear group lock, if present */
326 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
327 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
329 mutex_lock(&lli->lli_och_mutex);
330 if (fd->fd_lease_och != NULL) {
332 struct obd_client_handle *lease_och;
334 lease_och = fd->fd_lease_och;
335 fd->fd_lease_och = NULL;
336 mutex_unlock(&lli->lli_och_mutex);
338 /* Usually the lease is not released when the
339 * application crashed, we need to release here. */
340 rc = ll_lease_close(lease_och, inode, &lease_broken);
342 mutex_lock(&lli->lli_och_mutex);
344 CDEBUG_LIMIT(rc ? D_ERROR : D_INODE,
345 "Clean up lease "DFID" %d/%d\n",
346 PFID(&lli->lli_fid), rc, lease_broken);
349 if (fd->fd_och != NULL) {
350 struct obd_client_handle *och;
354 mutex_unlock(&lli->lli_och_mutex);
356 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
360 /* Let's see if we have good enough OPEN lock on the file and if
361 we can skip talking to MDS */
362 if (fd->fd_omode & FMODE_WRITE) {
364 LASSERT(lli->lli_open_fd_write_count);
365 lli->lli_open_fd_write_count--;
366 } else if (fd->fd_omode & FMODE_EXEC) {
368 LASSERT(lli->lli_open_fd_exec_count);
369 lli->lli_open_fd_exec_count--;
372 LASSERT(lli->lli_open_fd_read_count);
373 lli->lli_open_fd_read_count--;
375 mutex_unlock(&lli->lli_och_mutex);
377 /* LU-4398: do not cache write open lock if the file has exec bit */
378 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
379 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
380 LDLM_IBITS, &policy, lockmode, &lockh))
381 rc = ll_md_real_close(inode, fd->fd_omode);
384 file->private_data = NULL;
385 ll_file_data_put(fd);
390 /* While this returns an error code, fput() the caller does not, so we need
391 * to make every effort to clean up all of our state here. Also, applications
392 * rarely check close errors and even if an error is returned they will not
393 * re-try the close call.
395 int ll_file_release(struct inode *inode, struct file *file)
397 struct ll_file_data *fd;
398 struct ll_sb_info *sbi = ll_i2sbi(inode);
399 struct ll_inode_info *lli = ll_i2info(inode);
400 ktime_t kstart = ktime_get();
405 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
406 PFID(ll_inode2fid(inode)), inode);
408 fd = file->private_data;
411 /* The last ref on @file, maybe not the the owner pid of statahead,
412 * because parent and child process can share the same file handle. */
413 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
414 ll_deauthorize_statahead(inode, fd);
416 if (is_root_inode(inode)) {
417 file->private_data = NULL;
418 ll_file_data_put(fd);
422 pcc_file_release(inode, file);
424 if (!S_ISDIR(inode->i_mode)) {
425 if (lli->lli_clob != NULL)
426 lov_read_and_clear_async_rc(lli->lli_clob);
427 lli->lli_async_rc = 0;
430 lli->lli_close_fd_time = ktime_get();
432 rc = ll_md_close(inode, file);
434 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
435 libcfs_debug_dumplog();
438 if (!rc && !is_root_inode(inode))
439 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
440 ktime_us_delta(ktime_get(), kstart));
444 static inline int ll_dom_readpage(void *data, struct page *page)
446 /* since ll_dom_readpage is a page cache helper, it is safe to assume
447 * mapping and host pointers are set here
450 struct niobuf_local *lnb = data;
454 inode = page2inode(page);
456 kaddr = kmap_atomic(page);
457 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
458 if (lnb->lnb_len < PAGE_SIZE)
459 memset(kaddr + lnb->lnb_len, 0,
460 PAGE_SIZE - lnb->lnb_len);
461 flush_dcache_page(page);
462 SetPageUptodate(page);
463 kunmap_atomic(kaddr);
465 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
466 if (!llcrypt_has_encryption_key(inode))
467 CDEBUG(D_SEC, "no enc key for "DFID"\n",
468 PFID(ll_inode2fid(inode)));
470 unsigned int offs = 0;
472 while (offs < PAGE_SIZE) {
473 /* decrypt only if page is not empty */
474 if (memcmp(page_address(page) + offs,
475 page_address(ZERO_PAGE(0)),
476 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
479 rc = llcrypt_decrypt_pagecache_blocks(page,
480 LUSTRE_ENCRYPTION_UNIT_SIZE,
485 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
494 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
498 struct ll_inode_info *lli = ll_i2info(inode);
499 struct cl_object *obj = lli->lli_clob;
500 struct address_space *mapping = inode->i_mapping;
502 struct niobuf_remote *rnb;
503 struct mdt_body *body;
505 unsigned long index, start;
506 struct niobuf_local lnb;
515 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
519 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
520 if (rnb == NULL || rnb->rnb_len == 0)
523 /* LU-11595: Server may return whole file and that is OK always or
524 * it may return just file tail and its offset must be aligned with
525 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
526 * smaller then offset may be not aligned and that data is just ignored.
528 if (rnb->rnb_offset & ~PAGE_MASK)
531 /* Server returns whole file or just file tail if it fills in reply
532 * buffer, in both cases total size should be equal to the file size.
534 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
535 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
536 !(inode && IS_ENCRYPTED(inode))) {
537 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
538 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
539 rnb->rnb_len, body->mbo_dom_size);
543 env = cl_env_get(&refcheck);
546 io = vvp_env_thread_io(env);
548 io->ci_ignore_layout = 1;
549 rc = cl_io_init(env, io, CIT_MISC, obj);
553 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
554 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
556 data = (char *)rnb + sizeof(*rnb);
558 lnb.lnb_file_offset = rnb->rnb_offset;
559 start = lnb.lnb_file_offset >> PAGE_SHIFT;
561 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
562 lnb.lnb_page_offset = 0;
564 struct cl_page *page;
566 lnb.lnb_data = data + (index << PAGE_SHIFT);
567 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
568 if (lnb.lnb_len > PAGE_SIZE)
569 lnb.lnb_len = PAGE_SIZE;
571 vmpage = read_cache_page(mapping, index + start,
572 ll_dom_readpage, &lnb);
573 if (IS_ERR(vmpage)) {
574 CWARN("%s: cannot fill page %lu for "DFID
575 " with data: rc = %li\n",
576 ll_i2sbi(inode)->ll_fsname, index + start,
577 PFID(lu_object_fid(&obj->co_lu)),
582 if (vmpage->mapping == NULL) {
585 /* page was truncated */
588 /* attach VM page to CL page cache */
589 page = cl_page_find(env, obj, vmpage->index, vmpage,
592 ClearPageUptodate(vmpage);
597 cl_page_export(env, page, 1);
598 cl_page_put(env, page);
602 } while (rnb->rnb_len > (index << PAGE_SHIFT));
606 cl_env_put(env, &refcheck);
611 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
612 struct lookup_intent *itp)
614 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
615 struct dentry *parent = de->d_parent;
618 struct md_op_data *op_data;
619 struct ptlrpc_request *req = NULL;
623 LASSERT(parent != NULL);
624 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
626 /* if server supports open-by-fid, or file name is invalid, don't pack
627 * name in open request */
628 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
629 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
631 len = de->d_name.len;
632 name = kmalloc(len + 1, GFP_NOFS);
637 spin_lock(&de->d_lock);
638 if (len != de->d_name.len) {
639 spin_unlock(&de->d_lock);
643 memcpy(name, de->d_name.name, len);
645 spin_unlock(&de->d_lock);
647 if (!lu_name_is_valid_2(name, len)) {
653 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
654 name, len, 0, LUSTRE_OPC_OPEN, NULL);
655 if (IS_ERR(op_data)) {
657 RETURN(PTR_ERR(op_data));
659 op_data->op_data = lmm;
660 op_data->op_data_size = lmmsize;
662 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_OPEN_DELAY, cfs_fail_val);
664 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
665 &ll_md_blocking_ast, 0);
667 ll_finish_md_op_data(op_data);
669 /* reason for keep own exit path - don`t flood log
670 * with messages with -ESTALE errors.
672 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
673 it_open_error(DISP_OPEN_OPEN, itp))
675 ll_release_openhandle(de, itp);
679 if (it_disposition(itp, DISP_LOOKUP_NEG))
680 GOTO(out, rc = -ENOENT);
682 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
683 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
684 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
688 rc = ll_prep_inode(&de->d_inode, &req->rq_pill, NULL, itp);
690 if (!rc && itp->it_lock_mode) {
693 /* If we got a lock back and it has a LOOKUP bit set,
694 * make sure the dentry is marked as valid so we can find it.
695 * We don't need to care about actual hashing since other bits
696 * of kernel will deal with that later.
698 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
699 if (bits & MDS_INODELOCK_LOOKUP) {
700 d_lustre_revalidate(de);
701 ll_update_dir_depth(parent->d_inode, de->d_inode);
704 /* if DoM bit returned along with LAYOUT bit then there
705 * can be read-on-open data returned.
707 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
708 ll_dom_finish_open(de->d_inode, req);
712 ptlrpc_req_finished(req);
713 ll_intent_drop_lock(itp);
715 /* We did open by fid, but by the time we got to the server, the object
716 * disappeared. This is possible if the object was unlinked, but it's
717 * also possible if the object was unlinked by a rename. In the case
718 * of an object renamed over our existing one, we can't fail this open.
719 * O_CREAT also goes through this path if we had an existing dentry,
720 * and it's obviously wrong to return ENOENT for O_CREAT.
722 * Instead let's return -ESTALE, and the VFS will retry the open with
723 * LOOKUP_REVAL, which we catch in ll_revalidate_dentry and fail to
724 * revalidate, causing a lookup. This causes extra lookups in the case
725 * where we had a dentry in cache but the file is being unlinked and we
726 * lose the race with unlink, but this should be very rare.
734 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
735 struct obd_client_handle *och)
737 struct mdt_body *body;
739 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
740 och->och_open_handle = body->mbo_open_handle;
741 och->och_fid = body->mbo_fid1;
742 och->och_lease_handle.cookie = it->it_lock_handle;
743 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
744 och->och_flags = it->it_flags;
746 return md_set_open_replay_data(md_exp, och, it);
749 static int ll_local_open(struct file *file, struct lookup_intent *it,
750 struct ll_file_data *fd, struct obd_client_handle *och)
752 struct inode *inode = file_inode(file);
755 LASSERT(!file->private_data);
762 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
767 file->private_data = fd;
768 ll_readahead_init(inode, &fd->fd_ras);
769 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
774 void ll_track_file_opens(struct inode *inode)
776 struct ll_inode_info *lli = ll_i2info(inode);
777 struct ll_sb_info *sbi = ll_i2sbi(inode);
779 /* do not skew results with delays from never-opened inodes */
780 if (ktime_to_ns(lli->lli_close_fd_time))
781 ll_stats_ops_tally(sbi, LPROC_LL_INODE_OPCLTM,
782 ktime_us_delta(ktime_get(), lli->lli_close_fd_time));
784 if (ktime_after(ktime_get(),
785 ktime_add_ms(lli->lli_close_fd_time,
786 sbi->ll_oc_max_ms))) {
787 lli->lli_open_fd_count = 1;
788 lli->lli_close_fd_time = ns_to_ktime(0);
790 lli->lli_open_fd_count++;
793 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_OCOUNT,
794 lli->lli_open_fd_count);
797 /* Open a file, and (for the very first open) create objects on the OSTs at
798 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
799 * creation or open until ll_lov_setstripe() ioctl is called.
801 * If we already have the stripe MD locally then we don't request it in
802 * md_open(), by passing a lmm_size = 0.
804 * It is up to the application to ensure no other processes open this file
805 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
806 * used. We might be able to avoid races of that sort by getting lli_open_sem
807 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
808 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
810 int ll_file_open(struct inode *inode, struct file *file)
812 struct ll_inode_info *lli = ll_i2info(inode);
813 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
814 .it_flags = file->f_flags };
815 struct obd_client_handle **och_p = NULL;
816 __u64 *och_usecount = NULL;
817 struct ll_file_data *fd;
818 ktime_t kstart = ktime_get();
822 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
823 PFID(ll_inode2fid(inode)), inode, file->f_flags);
825 it = file->private_data; /* XXX: compat macro */
826 file->private_data = NULL; /* prevent ll_local_open assertion */
828 if (S_ISREG(inode->i_mode)) {
829 rc = ll_file_open_encrypt(inode, file);
831 if (it && it->it_disposition)
832 ll_release_openhandle(file_dentry(file), it);
833 GOTO(out_nofiledata, rc);
837 fd = ll_file_data_get();
839 GOTO(out_nofiledata, rc = -ENOMEM);
842 if (S_ISDIR(inode->i_mode))
843 ll_authorize_statahead(inode, fd);
845 ll_track_file_opens(inode);
846 if (is_root_inode(inode)) {
847 file->private_data = fd;
851 if (!it || !it->it_disposition) {
852 /* Convert f_flags into access mode. We cannot use file->f_mode,
853 * because everything but O_ACCMODE mask was stripped from
855 if ((oit.it_flags + 1) & O_ACCMODE)
857 if (file->f_flags & O_TRUNC)
858 oit.it_flags |= FMODE_WRITE;
860 /* kernel only call f_op->open in dentry_open. filp_open calls
861 * dentry_open after call to open_namei that checks permissions.
862 * Only nfsd_open call dentry_open directly without checking
863 * permissions and because of that this code below is safe.
865 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
866 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
868 /* We do not want O_EXCL here, presumably we opened the file
869 * already? XXX - NFS implications? */
870 oit.it_flags &= ~O_EXCL;
872 /* bug20584, if "it_flags" contains O_CREAT, the file will be
873 * created if necessary, then "IT_CREAT" should be set to keep
874 * consistent with it */
875 if (oit.it_flags & O_CREAT)
876 oit.it_op |= IT_CREAT;
882 /* Let's see if we have file open on MDS already. */
883 if (it->it_flags & FMODE_WRITE) {
884 och_p = &lli->lli_mds_write_och;
885 och_usecount = &lli->lli_open_fd_write_count;
886 } else if (it->it_flags & FMODE_EXEC) {
887 och_p = &lli->lli_mds_exec_och;
888 och_usecount = &lli->lli_open_fd_exec_count;
890 och_p = &lli->lli_mds_read_och;
891 och_usecount = &lli->lli_open_fd_read_count;
894 mutex_lock(&lli->lli_och_mutex);
895 if (*och_p) { /* Open handle is present */
896 if (it_disposition(it, DISP_OPEN_OPEN)) {
897 /* Well, there's extra open request that we do not need,
898 * let's close it somehow. This will decref request. */
899 rc = it_open_error(DISP_OPEN_OPEN, it);
901 mutex_unlock(&lli->lli_och_mutex);
902 GOTO(out_openerr, rc);
905 ll_release_openhandle(file_dentry(file), it);
909 rc = ll_local_open(file, it, fd, NULL);
912 mutex_unlock(&lli->lli_och_mutex);
913 GOTO(out_openerr, rc);
916 LASSERT(*och_usecount == 0);
917 if (!it->it_disposition) {
918 struct dentry *dentry = file_dentry(file);
919 struct ll_sb_info *sbi = ll_i2sbi(inode);
920 struct ll_dentry_data *ldd;
922 /* We cannot just request lock handle now, new ELC code
923 * means that one of other OPEN locks for this file
924 * could be cancelled, and since blocking ast handler
925 * would attempt to grab och_mutex as well, that would
926 * result in a deadlock
928 mutex_unlock(&lli->lli_och_mutex);
930 * Normally called under two situations:
932 * 2. A race/condition on MDS resulting in no open
933 * handle to be returned from LOOKUP|OPEN request,
934 * for example if the target entry was a symlink.
936 * In NFS path we know there's pathologic behavior
937 * so we always enable open lock caching when coming
938 * from there. It's detected by setting a flag in
941 * After reaching number of opens of this inode
942 * we always ask for an open lock on it to handle
943 * bad userspace actors that open and close files
944 * in a loop for absolutely no good reason
947 ldd = ll_d2d(dentry);
948 if (filename_is_volatile(dentry->d_name.name,
951 /* There really is nothing here, but this
952 * make this more readable I think.
953 * We do not want openlock for volatile
954 * files under any circumstances
956 } else if (ldd && ldd->lld_nfs_dentry) {
957 /* NFS path. This also happens to catch
958 * open by fh files I guess
960 it->it_flags |= MDS_OPEN_LOCK;
961 /* clear the flag for future lookups */
962 ldd->lld_nfs_dentry = 0;
963 } else if (sbi->ll_oc_thrsh_count > 0) {
964 /* Take MDS_OPEN_LOCK with many opens */
965 if (lli->lli_open_fd_count >=
966 sbi->ll_oc_thrsh_count)
967 it->it_flags |= MDS_OPEN_LOCK;
969 /* If this is open after we just closed */
970 else if (ktime_before(ktime_get(),
971 ktime_add_ms(lli->lli_close_fd_time,
972 sbi->ll_oc_thrsh_ms)))
973 it->it_flags |= MDS_OPEN_LOCK;
977 * Always specify MDS_OPEN_BY_FID because we don't want
978 * to get file with different fid.
980 it->it_flags |= MDS_OPEN_BY_FID;
981 rc = ll_intent_file_open(dentry, NULL, 0, it);
983 GOTO(out_openerr, rc);
987 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
989 GOTO(out_och_free, rc = -ENOMEM);
993 /* md_intent_lock() didn't get a request ref if there was an
994 * open error, so don't do cleanup on the request here
996 /* XXX (green): Should not we bail out on any error here, not
997 * just open error? */
998 rc = it_open_error(DISP_OPEN_OPEN, it);
1000 GOTO(out_och_free, rc);
1002 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
1003 "inode %p: disposition %x, status %d\n", inode,
1004 it_disposition(it, ~0), it->it_status);
1006 rc = ll_local_open(file, it, fd, *och_p);
1008 GOTO(out_och_free, rc);
1011 rc = pcc_file_open(inode, file);
1013 GOTO(out_och_free, rc);
1015 mutex_unlock(&lli->lli_och_mutex);
1019 /* Must do this outside lli_och_mutex lock to prevent deadlock where
1020 different kind of OPEN lock for this same inode gets cancelled
1021 by ldlm_cancel_lru */
1022 if (!S_ISREG(inode->i_mode))
1023 GOTO(out_och_free, rc);
1024 cl_lov_delay_create_clear(&file->f_flags);
1025 GOTO(out_och_free, rc);
1029 if (och_p && *och_p) {
1030 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
1031 *och_p = NULL; /* OBD_FREE writes some magic there */
1034 mutex_unlock(&lli->lli_och_mutex);
1037 if (lli->lli_opendir_key == fd)
1038 ll_deauthorize_statahead(inode, fd);
1041 ll_file_data_put(fd);
1043 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
1044 ktime_us_delta(ktime_get(), kstart));
1048 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
1049 ptlrpc_req_finished(it->it_request);
1050 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1056 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
1057 struct ldlm_lock_desc *desc, void *data, int flag)
1060 struct lustre_handle lockh;
1064 case LDLM_CB_BLOCKING:
1065 ldlm_lock2handle(lock, &lockh);
1066 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
1068 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1072 case LDLM_CB_CANCELING:
1080 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1081 * and save it as fd->fd_och so as to force client to reopen the file even
1082 * if it has an open lock in cache already.
1084 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1085 struct lustre_handle *old_open_handle)
1087 struct ll_inode_info *lli = ll_i2info(inode);
1088 struct ll_file_data *fd = file->private_data;
1089 struct obd_client_handle **och_p;
1090 __u64 *och_usecount;
1094 /* Get the openhandle of the file */
1095 mutex_lock(&lli->lli_och_mutex);
1096 if (fd->fd_lease_och != NULL)
1097 GOTO(out_unlock, rc = -EBUSY);
1099 if (fd->fd_och == NULL) {
1100 if (file->f_mode & FMODE_WRITE) {
1101 LASSERT(lli->lli_mds_write_och != NULL);
1102 och_p = &lli->lli_mds_write_och;
1103 och_usecount = &lli->lli_open_fd_write_count;
1105 LASSERT(lli->lli_mds_read_och != NULL);
1106 och_p = &lli->lli_mds_read_och;
1107 och_usecount = &lli->lli_open_fd_read_count;
1110 if (*och_usecount > 1)
1111 GOTO(out_unlock, rc = -EBUSY);
1113 fd->fd_och = *och_p;
1118 *old_open_handle = fd->fd_och->och_open_handle;
1122 mutex_unlock(&lli->lli_och_mutex);
1127 * Release ownership on lli_mds_*_och when putting back a file lease.
1129 static int ll_lease_och_release(struct inode *inode, struct file *file)
1131 struct ll_inode_info *lli = ll_i2info(inode);
1132 struct ll_file_data *fd = file->private_data;
1133 struct obd_client_handle **och_p;
1134 struct obd_client_handle *old_och = NULL;
1135 __u64 *och_usecount;
1139 mutex_lock(&lli->lli_och_mutex);
1140 if (file->f_mode & FMODE_WRITE) {
1141 och_p = &lli->lli_mds_write_och;
1142 och_usecount = &lli->lli_open_fd_write_count;
1144 och_p = &lli->lli_mds_read_och;
1145 och_usecount = &lli->lli_open_fd_read_count;
1148 /* The file may have been open by another process (broken lease) so
1149 * *och_p is not NULL. In this case we should simply increase usecount
1152 if (*och_p != NULL) {
1153 old_och = fd->fd_och;
1156 *och_p = fd->fd_och;
1160 mutex_unlock(&lli->lli_och_mutex);
1162 if (old_och != NULL)
1163 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1169 * Acquire a lease and open the file.
1171 static struct obd_client_handle *
1172 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1175 struct lookup_intent it = { .it_op = IT_OPEN };
1176 struct ll_sb_info *sbi = ll_i2sbi(inode);
1177 struct md_op_data *op_data;
1178 struct ptlrpc_request *req = NULL;
1179 struct lustre_handle old_open_handle = { 0 };
1180 struct obd_client_handle *och = NULL;
1185 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1186 RETURN(ERR_PTR(-EINVAL));
1189 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1190 RETURN(ERR_PTR(-EPERM));
1192 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1194 RETURN(ERR_PTR(rc));
1199 RETURN(ERR_PTR(-ENOMEM));
1201 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1202 LUSTRE_OPC_ANY, NULL);
1203 if (IS_ERR(op_data))
1204 GOTO(out, rc = PTR_ERR(op_data));
1206 /* To tell the MDT this openhandle is from the same owner */
1207 op_data->op_open_handle = old_open_handle;
1209 it.it_flags = fmode | open_flags;
1210 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1211 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1212 &ll_md_blocking_lease_ast,
1213 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1214 * it can be cancelled which may mislead applications that the lease is
1216 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1217 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1218 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1219 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1220 ll_finish_md_op_data(op_data);
1221 ptlrpc_req_finished(req);
1223 GOTO(out_release_it, rc);
1225 if (it_disposition(&it, DISP_LOOKUP_NEG))
1226 GOTO(out_release_it, rc = -ENOENT);
1228 rc = it_open_error(DISP_OPEN_OPEN, &it);
1230 GOTO(out_release_it, rc);
1232 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1233 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1235 GOTO(out_release_it, rc);
1237 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1238 GOTO(out_close, rc = -EOPNOTSUPP);
1240 /* already get lease, handle lease lock */
1241 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1242 if (!it.it_lock_mode ||
1243 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1244 /* open lock must return for lease */
1245 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1246 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1248 GOTO(out_close, rc = -EPROTO);
1251 ll_intent_release(&it);
1255 /* Cancel open lock */
1256 if (it.it_lock_mode != 0) {
1257 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1259 it.it_lock_mode = 0;
1260 och->och_lease_handle.cookie = 0ULL;
1262 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1264 CERROR("%s: error closing file "DFID": %d\n",
1265 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1266 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1268 ll_intent_release(&it);
1272 RETURN(ERR_PTR(rc));
1276 * Check whether a layout swap can be done between two inodes.
1278 * \param[in] inode1 First inode to check
1279 * \param[in] inode2 Second inode to check
1281 * \retval 0 on success, layout swap can be performed between both inodes
1282 * \retval negative error code if requirements are not met
1284 static int ll_check_swap_layouts_validity(struct inode *inode1,
1285 struct inode *inode2)
1287 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1290 if (inode_permission(&init_user_ns, inode1, MAY_WRITE) ||
1291 inode_permission(&init_user_ns, inode2, MAY_WRITE))
1294 if (inode1->i_sb != inode2->i_sb)
1300 static int ll_swap_layouts_close(struct obd_client_handle *och,
1301 struct inode *inode, struct inode *inode2)
1303 const struct lu_fid *fid1 = ll_inode2fid(inode);
1304 const struct lu_fid *fid2;
1308 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1309 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1311 rc = ll_check_swap_layouts_validity(inode, inode2);
1313 GOTO(out_free_och, rc);
1315 /* We now know that inode2 is a lustre inode */
1316 fid2 = ll_inode2fid(inode2);
1318 rc = lu_fid_cmp(fid1, fid2);
1320 GOTO(out_free_och, rc = -EINVAL);
1322 /* Close the file and {swap,merge} layouts between inode & inode2.
1323 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1324 * because we still need it to pack l_remote_handle to MDT. */
1325 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1328 och = NULL; /* freed in ll_close_inode_openhandle() */
1338 * Release lease and close the file.
1339 * It will check if the lease has ever broken.
1341 static int ll_lease_close_intent(struct obd_client_handle *och,
1342 struct inode *inode,
1343 bool *lease_broken, enum mds_op_bias bias,
1346 struct ldlm_lock *lock;
1347 bool cancelled = true;
1351 lock = ldlm_handle2lock(&och->och_lease_handle);
1353 lock_res_and_lock(lock);
1354 cancelled = ldlm_is_cancel(lock);
1355 unlock_res_and_lock(lock);
1356 LDLM_LOCK_PUT(lock);
1359 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1360 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1362 if (lease_broken != NULL)
1363 *lease_broken = cancelled;
1365 if (!cancelled && !bias)
1366 ldlm_cli_cancel(&och->och_lease_handle, 0);
1368 if (cancelled) { /* no need to excute intent */
1373 rc = ll_close_inode_openhandle(inode, och, bias, data);
1377 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1380 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1384 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1386 static int ll_lease_file_resync(struct obd_client_handle *och,
1387 struct inode *inode, unsigned long arg)
1389 struct ll_sb_info *sbi = ll_i2sbi(inode);
1390 struct md_op_data *op_data;
1391 struct ll_ioc_lease_id ioc;
1392 __u64 data_version_unused;
1396 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1397 LUSTRE_OPC_ANY, NULL);
1398 if (IS_ERR(op_data))
1399 RETURN(PTR_ERR(op_data));
1401 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1405 /* before starting file resync, it's necessary to clean up page cache
1406 * in client memory, otherwise once the layout version is increased,
1407 * writing back cached data will be denied the OSTs. */
1408 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1412 op_data->op_lease_handle = och->och_lease_handle;
1413 op_data->op_mirror_id = ioc.lil_mirror_id;
1414 rc = md_file_resync(sbi->ll_md_exp, op_data);
1420 ll_finish_md_op_data(op_data);
1424 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1426 struct ll_inode_info *lli = ll_i2info(inode);
1427 struct cl_object *obj = lli->lli_clob;
1428 struct cl_attr *attr = vvp_env_thread_attr(env);
1436 ll_inode_size_lock(inode);
1438 /* Merge timestamps the most recently obtained from MDS with
1439 * timestamps obtained from OSTs.
1441 * Do not overwrite atime of inode because it may be refreshed
1442 * by file_accessed() function. If the read was served by cache
1443 * data, there is no RPC to be sent so that atime may not be
1444 * transferred to OSTs at all. MDT only updates atime at close time
1445 * if it's at least 'mdd.*.atime_diff' older.
1446 * All in all, the atime in Lustre does not strictly comply with
1447 * POSIX. Solving this problem needs to send an RPC to MDT for each
1448 * read, this will hurt performance.
1450 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1451 inode->i_atime.tv_sec < lli->lli_atime)
1452 inode->i_atime.tv_sec = lli->lli_atime;
1454 inode->i_mtime.tv_sec = lli->lli_mtime;
1455 inode->i_ctime.tv_sec = lli->lli_ctime;
1457 mtime = inode->i_mtime.tv_sec;
1458 atime = inode->i_atime.tv_sec;
1459 ctime = inode->i_ctime.tv_sec;
1461 cl_object_attr_lock(obj);
1462 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1465 rc = cl_object_attr_get(env, obj, attr);
1466 cl_object_attr_unlock(obj);
1469 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1471 if (atime < attr->cat_atime)
1472 atime = attr->cat_atime;
1474 if (ctime < attr->cat_ctime)
1475 ctime = attr->cat_ctime;
1477 if (mtime < attr->cat_mtime)
1478 mtime = attr->cat_mtime;
1480 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1481 PFID(&lli->lli_fid), attr->cat_size);
1483 if (llcrypt_require_key(inode) == -ENOKEY) {
1484 /* Without the key, round up encrypted file size to next
1485 * LUSTRE_ENCRYPTION_UNIT_SIZE. Clear text size is put in
1486 * lli_lazysize for proper file size setting at close time.
1488 lli->lli_attr_valid |= OBD_MD_FLLAZYSIZE;
1489 lli->lli_lazysize = attr->cat_size;
1490 attr->cat_size = round_up(attr->cat_size,
1491 LUSTRE_ENCRYPTION_UNIT_SIZE);
1493 i_size_write(inode, attr->cat_size);
1494 inode->i_blocks = attr->cat_blocks;
1496 inode->i_mtime.tv_sec = mtime;
1497 inode->i_atime.tv_sec = atime;
1498 inode->i_ctime.tv_sec = ctime;
1501 ll_inode_size_unlock(inode);
1507 * Set designated mirror for I/O.
1509 * So far only read, write, and truncated can support to issue I/O to
1510 * designated mirror.
1512 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1514 struct ll_file_data *fd = file->private_data;
1516 /* clear layout version for generic(non-resync) I/O in case it carries
1517 * stale layout version due to I/O restart */
1518 io->ci_layout_version = 0;
1520 /* FLR: disable non-delay for designated mirror I/O because obviously
1521 * only one mirror is available */
1522 if (fd->fd_designated_mirror > 0) {
1524 io->ci_designated_mirror = fd->fd_designated_mirror;
1525 io->ci_layout_version = fd->fd_layout_version;
1528 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1529 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1532 static bool file_is_noatime(const struct file *file)
1534 const struct vfsmount *mnt = file->f_path.mnt;
1535 const struct inode *inode = file_inode((struct file *)file);
1537 /* Adapted from file_accessed() and touch_atime().*/
1538 if (file->f_flags & O_NOATIME)
1541 if (inode->i_flags & S_NOATIME)
1544 if (IS_NOATIME(inode))
1547 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1550 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1553 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1559 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1560 struct vvp_io_args *args)
1562 struct inode *inode = file_inode(file);
1563 struct ll_file_data *fd = file->private_data;
1565 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1566 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1568 if (iot == CIT_WRITE) {
1569 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1570 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1571 file->f_flags & O_DIRECT ||
1573 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1574 io->u.ci_wr.wr_sync |= !!(args &&
1575 (args->u.normal.via_iocb->ki_flags &
1580 io->ci_obj = ll_i2info(inode)->lli_clob;
1581 io->ci_lockreq = CILR_MAYBE;
1582 if (ll_file_nolock(file)) {
1583 io->ci_lockreq = CILR_NEVER;
1584 io->ci_no_srvlock = 1;
1585 } else if (file->f_flags & O_APPEND) {
1586 io->ci_lockreq = CILR_MANDATORY;
1588 io->ci_noatime = file_is_noatime(file);
1589 io->ci_async_readahead = false;
1591 /* FLR: only use non-delay I/O for read as there is only one
1592 * avaliable mirror for write. */
1593 io->ci_ndelay = !(iot == CIT_WRITE);
1595 ll_io_set_mirror(io, file);
1598 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1601 struct ll_inode_info *lli = ll_i2info(inode);
1602 struct ll_sb_info *sbi = ll_i2sbi(inode);
1603 enum obd_heat_type sample_type;
1604 enum obd_heat_type iobyte_type;
1605 __u64 now = ktime_get_real_seconds();
1607 if (!ll_sbi_has_file_heat(sbi) ||
1608 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1611 if (iot == CIT_READ) {
1612 sample_type = OBD_HEAT_READSAMPLE;
1613 iobyte_type = OBD_HEAT_READBYTE;
1614 } else if (iot == CIT_WRITE) {
1615 sample_type = OBD_HEAT_WRITESAMPLE;
1616 iobyte_type = OBD_HEAT_WRITEBYTE;
1621 spin_lock(&lli->lli_heat_lock);
1622 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1623 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1624 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1625 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1626 spin_unlock(&lli->lli_heat_lock);
1630 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1631 struct file *file, enum cl_io_type iot,
1632 loff_t *ppos, size_t count)
1634 struct vvp_io *vio = vvp_env_io(env);
1635 struct inode *inode = file_inode(file);
1636 struct ll_inode_info *lli = ll_i2info(inode);
1637 struct ll_sb_info *sbi = ll_i2sbi(inode);
1638 struct ll_file_data *fd = file->private_data;
1639 struct range_lock range;
1640 bool range_locked = false;
1645 unsigned int retried = 0, dio_lock = 0;
1646 bool is_aio = false;
1647 bool is_parallel_dio = false;
1648 struct cl_dio_aio *ci_dio_aio = NULL;
1650 bool partial_io = false;
1651 size_t max_io_pages, max_cached_pages;
1655 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1656 file_dentry(file)->d_name.name,
1657 iot == CIT_READ ? "read" : "write", *ppos, count);
1659 max_io_pages = PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT;
1660 max_cached_pages = sbi->ll_cache->ccc_lru_max;
1661 if (max_io_pages > (max_cached_pages >> 2))
1662 max_io_pages = max_cached_pages >> 2;
1664 io = vvp_env_thread_io(env);
1665 if (file->f_flags & O_DIRECT) {
1666 if (file->f_flags & O_APPEND)
1668 if (!is_sync_kiocb(args->u.normal.via_iocb))
1671 /* the kernel does not support AIO on pipes, and parallel DIO
1672 * uses part of the AIO path, so we must not do parallel dio
1675 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1678 if (!ll_sbi_has_parallel_dio(sbi))
1679 is_parallel_dio = false;
1681 ci_dio_aio = cl_dio_aio_alloc(args->u.normal.via_iocb,
1682 ll_i2info(inode)->lli_clob, is_aio);
1684 GOTO(out, rc = -ENOMEM);
1689 * IO block size need be aware of cached page limit, otherwise
1690 * if we have small max_cached_mb but large block IO issued, io
1691 * could not be finished and blocked whole client.
1693 if (file->f_flags & O_DIRECT)
1696 per_bytes = min(max_io_pages << PAGE_SHIFT, count);
1697 partial_io = per_bytes < count;
1698 io = vvp_env_thread_io(env);
1699 ll_io_init(io, file, iot, args);
1700 io->ci_dio_aio = ci_dio_aio;
1701 io->ci_dio_lock = dio_lock;
1702 io->ci_ndelay_tried = retried;
1703 io->ci_parallel_dio = is_parallel_dio;
1705 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1706 if (file->f_flags & O_APPEND)
1707 range_lock_init(&range, 0, LUSTRE_EOF);
1709 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1711 vio->vui_fd = file->private_data;
1712 vio->vui_iter = args->u.normal.via_iter;
1713 vio->vui_iocb = args->u.normal.via_iocb;
1714 /* Direct IO reads must also take range lock,
1715 * or multiple reads will try to work on the same pages
1716 * See LU-6227 for details.
1718 if (((iot == CIT_WRITE) ||
1719 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1720 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1721 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1723 rc = range_lock(&lli->lli_write_tree, &range);
1727 range_locked = true;
1730 ll_cl_add(inode, env, io, LCC_RW);
1731 rc = cl_io_loop(env, io);
1732 ll_cl_remove(inode, env);
1734 /* cl_io_rw_init() handled IO */
1738 if (io->ci_dio_aio && !is_aio) {
1739 struct cl_sync_io *anchor = &io->ci_dio_aio->cda_sync;
1741 /* for dio, EIOCBQUEUED is an implementation detail,
1742 * and we don't return it to userspace
1744 if (rc == -EIOCBQUEUED)
1747 /* N/B: parallel DIO may be disabled during i/o submission;
1748 * if that occurs, I/O shifts to sync, so it's all resolved
1749 * before we get here, and this wait call completes
1752 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1758 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1760 range_unlock(&lli->lli_write_tree, &range);
1761 range_locked = false;
1765 * In order to move forward AIO, ci_nob was increased,
1766 * but that doesn't mean io have been finished, it just
1767 * means io have been submited, we will always return
1768 * EIOCBQUEUED to the caller, So we could only return
1769 * number of bytes in non-AIO case.
1771 if (io->ci_nob > 0) {
1774 result += io->ci_nob;
1775 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1780 count -= io->ci_nob;
1782 /* prepare IO restart */
1784 args->u.normal.via_iter = vio->vui_iter;
1788 * Reexpand iov count because it was zero
1791 iov_iter_reexpand(vio->vui_iter, count);
1792 if (per_bytes == io->ci_nob)
1793 io->ci_need_restart = 1;
1797 cl_io_fini(env, io);
1800 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1801 file->f_path.dentry->d_name.name,
1802 iot, rc, result, io->ci_need_restart);
1804 if ((rc == 0 || rc == -ENODATA || rc == -ENOLCK) &&
1805 count > 0 && io->ci_need_restart) {
1807 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1808 file_dentry(file)->d_name.name,
1809 iot == CIT_READ ? "read" : "write",
1810 *ppos, count, result, rc);
1811 /* preserve the tried count for FLR */
1812 retried = io->ci_ndelay_tried;
1813 dio_lock = io->ci_dio_lock;
1817 if (io->ci_dio_aio) {
1819 * VFS will call aio_complete() if no -EIOCBQUEUED
1820 * is returned for AIO, so we can not call aio_complete()
1823 * NB: This is safe because the atomic_dec_and_lock in
1824 * cl_sync_io_init has implicit memory barriers, so this will
1825 * be seen by whichever thread completes the DIO/AIO, even if
1828 if (rc != -EIOCBQUEUED)
1829 io->ci_dio_aio->cda_no_aio_complete = 1;
1831 * Drop one extra reference so that end_io() could be
1832 * called for this IO context, we could call it after
1833 * we make sure all AIO requests have been proceed.
1835 cl_sync_io_note(env, &io->ci_dio_aio->cda_sync,
1836 rc == -EIOCBQUEUED ? 0 : rc);
1838 LASSERT(io->ci_dio_aio->cda_creator_free);
1839 cl_dio_aio_free(env, io->ci_dio_aio);
1840 io->ci_dio_aio = NULL;
1844 if (iot == CIT_READ) {
1846 ll_stats_ops_tally(ll_i2sbi(inode),
1847 LPROC_LL_READ_BYTES, result);
1848 } else if (iot == CIT_WRITE) {
1850 ll_stats_ops_tally(ll_i2sbi(inode),
1851 LPROC_LL_WRITE_BYTES, result);
1852 fd->fd_write_failed = false;
1853 } else if (result == 0 && rc == 0) {
1856 fd->fd_write_failed = true;
1858 fd->fd_write_failed = false;
1859 } else if (rc != -ERESTARTSYS) {
1860 fd->fd_write_failed = true;
1864 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1866 ll_heat_add(inode, iot, result);
1868 RETURN(result > 0 ? result : rc);
1872 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1873 * especially for small I/O.
1875 * To serve a read request, CLIO has to create and initialize a cl_io and
1876 * then request DLM lock. This has turned out to have siginificant overhead
1877 * and affects the performance of small I/O dramatically.
1879 * It's not necessary to create a cl_io for each I/O. Under the help of read
1880 * ahead, most of the pages being read are already in memory cache and we can
1881 * read those pages directly because if the pages exist, the corresponding DLM
1882 * lock must exist so that page content must be valid.
1884 * In fast read implementation, the llite speculatively finds and reads pages
1885 * in memory cache. There are three scenarios for fast read:
1886 * - If the page exists and is uptodate, kernel VM will provide the data and
1887 * CLIO won't be intervened;
1888 * - If the page was brought into memory by read ahead, it will be exported
1889 * and read ahead parameters will be updated;
1890 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1891 * it will go back and invoke normal read, i.e., a cl_io will be created
1892 * and DLM lock will be requested.
1894 * POSIX compliance: posix standard states that read is intended to be atomic.
1895 * Lustre read implementation is in line with Linux kernel read implementation
1896 * and neither of them complies with POSIX standard in this matter. Fast read
1897 * doesn't make the situation worse on single node but it may interleave write
1898 * results from multiple nodes due to short read handling in ll_file_aio_read().
1900 * \param env - lu_env
1901 * \param iocb - kiocb from kernel
1902 * \param iter - user space buffers where the data will be copied
1904 * \retval - number of bytes have been read, or error code if error occurred.
1907 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1911 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1914 /* NB: we can't do direct IO for fast read because it will need a lock
1915 * to make IO engine happy. */
1916 if (iocb->ki_filp->f_flags & O_DIRECT)
1919 result = generic_file_read_iter(iocb, iter);
1921 /* If the first page is not in cache, generic_file_aio_read() will be
1922 * returned with -ENODATA.
1923 * See corresponding code in ll_readpage(). */
1924 if (result == -ENODATA)
1928 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1929 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1930 LPROC_LL_READ_BYTES, result);
1937 * Read from a file (through the page cache).
1939 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1942 struct vvp_io_args *args;
1943 struct file *file = iocb->ki_filp;
1947 ktime_t kstart = ktime_get();
1952 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
1953 file_dentry(file)->d_name.name,
1954 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
1955 iov_iter_count(to));
1957 if (!iov_iter_count(to))
1961 * Currently when PCC read failed, we do not fall back to the
1962 * normal read path, just return the error.
1963 * The resaon is that: for RW-PCC, the file data may be modified
1964 * in the PCC and inconsistent with the data on OSTs (or file
1965 * data has been removed from the Lustre file system), at this
1966 * time, fallback to the normal read path may read the wrong
1968 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1969 * path: read data from data copy on OSTs.
1971 result = pcc_file_read_iter(iocb, to, &cached);
1975 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
1977 result = ll_do_fast_read(iocb, to);
1978 if (result < 0 || iov_iter_count(to) == 0)
1981 env = cl_env_get(&refcheck);
1983 RETURN(PTR_ERR(env));
1985 args = ll_env_args(env);
1986 args->u.normal.via_iter = to;
1987 args->u.normal.via_iocb = iocb;
1989 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
1990 &iocb->ki_pos, iov_iter_count(to));
1993 else if (result == 0)
1996 cl_env_put(env, &refcheck);
1999 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2000 file->private_data, iocb->ki_pos, result,
2002 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
2003 ktime_us_delta(ktime_get(), kstart));
2010 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
2011 * If a page is already in the page cache and dirty (and some other things -
2012 * See ll_tiny_write_begin for the instantiation of these rules), then we can
2013 * write to it without doing a full I/O, because Lustre already knows about it
2014 * and will write it out. This saves a lot of processing time.
2016 * All writes here are within one page, so exclusion is handled by the page
2017 * lock on the vm page. We do not do tiny writes for writes which touch
2018 * multiple pages because it's very unlikely multiple sequential pages are
2019 * are already dirty.
2021 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
2022 * and are unlikely to be to already dirty pages.
2024 * Attribute updates are important here, we do them in ll_tiny_write_end.
2026 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
2028 ssize_t count = iov_iter_count(iter);
2029 struct file *file = iocb->ki_filp;
2030 struct inode *inode = file_inode(file);
2031 bool lock_inode = !IS_NOSEC(inode);
2036 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2037 * of function for why.
2039 if (count >= PAGE_SIZE ||
2040 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2043 if (unlikely(lock_inode))
2045 result = __generic_file_write_iter(iocb, iter);
2047 if (unlikely(lock_inode))
2048 inode_unlock(inode);
2050 /* If the page is not already dirty, ll_tiny_write_begin returns
2051 * -ENODATA. We continue on to normal write.
2053 if (result == -ENODATA)
2057 ll_heat_add(inode, CIT_WRITE, result);
2058 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2061 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2067 * Write to a file (through the page cache).
2069 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2071 struct vvp_io_args *args;
2073 ssize_t rc_tiny = 0, rc_normal;
2074 struct file *file = iocb->ki_filp;
2077 ktime_t kstart = ktime_get();
2082 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2083 file_dentry(file)->d_name.name,
2084 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2085 iov_iter_count(from));
2087 if (!iov_iter_count(from))
2088 GOTO(out, rc_normal = 0);
2091 * When PCC write failed, we usually do not fall back to the normal
2092 * write path, just return the error. But there is a special case when
2093 * returned error code is -ENOSPC due to running out of space on PCC HSM
2094 * bakcend. At this time, it will fall back to normal I/O path and
2095 * retry the I/O. As the file is in HSM released state, it will restore
2096 * the file data to OSTs first and redo the write again. And the
2097 * restore process will revoke the layout lock and detach the file
2098 * from PCC cache automatically.
2100 result = pcc_file_write_iter(iocb, from, &cached);
2101 if (cached && result != -ENOSPC && result != -EDQUOT)
2102 GOTO(out, rc_normal = result);
2104 /* NB: we can't do direct IO for tiny writes because they use the page
2105 * cache, we can't do sync writes because tiny writes can't flush
2106 * pages, and we can't do append writes because we can't guarantee the
2107 * required DLM locks are held to protect file size.
2109 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2110 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
2111 rc_tiny = ll_do_tiny_write(iocb, from);
2113 /* In case of error, go on and try normal write - Only stop if tiny
2114 * write completed I/O.
2116 if (iov_iter_count(from) == 0)
2117 GOTO(out, rc_normal = rc_tiny);
2119 env = cl_env_get(&refcheck);
2121 RETURN(PTR_ERR(env));
2123 args = ll_env_args(env);
2124 args->u.normal.via_iter = from;
2125 args->u.normal.via_iocb = iocb;
2127 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2128 &iocb->ki_pos, iov_iter_count(from));
2130 /* On success, combine bytes written. */
2131 if (rc_tiny >= 0 && rc_normal > 0)
2132 rc_normal += rc_tiny;
2133 /* On error, only return error from normal write if tiny write did not
2134 * write any bytes. Otherwise return bytes written by tiny write.
2136 else if (rc_tiny > 0)
2137 rc_normal = rc_tiny;
2139 cl_env_put(env, &refcheck);
2141 if (rc_normal > 0) {
2142 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2143 file->private_data, iocb->ki_pos,
2145 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2146 ktime_us_delta(ktime_get(), kstart));
2152 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2154 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2156 static int ll_file_get_iov_count(const struct iovec *iov,
2157 unsigned long *nr_segs, size_t *count,
2163 for (seg = 0; seg < *nr_segs; seg++) {
2164 const struct iovec *iv = &iov[seg];
2167 * If any segment has a negative length, or the cumulative
2168 * length ever wraps negative then return -EINVAL.
2171 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2173 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2178 cnt -= iv->iov_len; /* This segment is no good */
2185 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2186 unsigned long nr_segs, loff_t pos)
2193 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2200 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2201 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2202 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2203 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2204 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2206 result = ll_file_read_iter(iocb, &to);
2211 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2214 struct iovec iov = { .iov_base = buf, .iov_len = count };
2223 init_sync_kiocb(&kiocb, file);
2224 kiocb.ki_pos = *ppos;
2225 #ifdef HAVE_KIOCB_KI_LEFT
2226 kiocb.ki_left = count;
2227 #elif defined(HAVE_KI_NBYTES)
2228 kiocb.i_nbytes = count;
2231 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2232 *ppos = kiocb.ki_pos;
2238 * Write to a file (through the page cache).
2241 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2242 unsigned long nr_segs, loff_t pos)
2244 struct iov_iter from;
2249 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2256 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2257 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2258 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2259 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2260 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2262 result = ll_file_write_iter(iocb, &from);
2267 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2268 size_t count, loff_t *ppos)
2270 struct iovec iov = { .iov_base = (void __user *)buf,
2280 init_sync_kiocb(&kiocb, file);
2281 kiocb.ki_pos = *ppos;
2282 #ifdef HAVE_KIOCB_KI_LEFT
2283 kiocb.ki_left = count;
2284 #elif defined(HAVE_KI_NBYTES)
2285 kiocb.ki_nbytes = count;
2288 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2289 *ppos = kiocb.ki_pos;
2293 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2295 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2296 __u64 flags, struct lov_user_md *lum, int lum_size)
2298 struct lookup_intent oit = {
2300 .it_flags = flags | MDS_OPEN_BY_FID,
2305 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2306 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2307 /* this code will only exist for big-endian systems */
2308 lustre_swab_lov_user_md(lum, 0);
2311 ll_inode_size_lock(inode);
2312 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2314 GOTO(out_unlock, rc);
2316 ll_release_openhandle(dentry, &oit);
2319 ll_inode_size_unlock(inode);
2320 ll_intent_release(&oit);
2325 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2326 struct lov_mds_md **lmmp, int *lmm_size,
2327 struct ptlrpc_request **request)
2329 struct ll_sb_info *sbi = ll_i2sbi(inode);
2330 struct mdt_body *body;
2331 struct lov_mds_md *lmm = NULL;
2332 struct ptlrpc_request *req = NULL;
2333 struct md_op_data *op_data;
2338 rc = ll_get_default_mdsize(sbi, &lmmsize);
2342 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2343 strlen(filename), lmmsize,
2344 LUSTRE_OPC_ANY, NULL);
2345 if (IS_ERR(op_data))
2346 RETURN(PTR_ERR(op_data));
2348 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2349 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2350 ll_finish_md_op_data(op_data);
2352 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2357 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2358 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2360 lmmsize = body->mbo_eadatasize;
2362 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2364 GOTO(out, rc = -ENODATA);
2366 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2367 LASSERT(lmm != NULL);
2369 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2370 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2371 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2372 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2373 GOTO(out, rc = -EPROTO);
2376 * This is coming from the MDS, so is probably in
2377 * little endian. We convert it to host endian before
2378 * passing it to userspace.
2380 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2381 int stripe_count = 0;
2383 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2384 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2385 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2386 if (le32_to_cpu(lmm->lmm_pattern) &
2387 LOV_PATTERN_F_RELEASED)
2389 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2391 /* if function called for directory - we should
2392 * avoid swab not existent lsm objects
2394 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2395 S_ISREG(body->mbo_mode))
2396 lustre_swab_lov_user_md_objects(
2397 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2399 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2400 S_ISREG(body->mbo_mode))
2401 lustre_swab_lov_user_md_objects(
2402 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2404 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2405 lustre_swab_lov_comp_md_v1(
2406 (struct lov_comp_md_v1 *)lmm);
2410 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2411 struct lov_comp_md_v1 *comp_v1 = NULL;
2412 struct lov_comp_md_entry_v1 *ent;
2413 struct lov_user_md_v1 *v1;
2417 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2418 /* Dump the striping information */
2419 for (; i < comp_v1->lcm_entry_count; i++) {
2420 ent = &comp_v1->lcm_entries[i];
2421 off = ent->lcme_offset;
2422 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2424 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2425 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2429 * Return valid stripe_count and stripe_size instead of 0 for
2430 * DoM files to avoid divide-by-zero for older userspace that
2431 * calls this ioctl, e.g. lustre ADIO driver.
2433 if (lmm->lmm_stripe_count == 0)
2434 lmm->lmm_stripe_count = 1;
2435 if (lmm->lmm_stripe_size == 0) {
2436 /* Since the first component of the file data is placed
2437 * on the MDT for faster access, the stripe_size of the
2438 * second one is always that applications which are
2441 if (lmm->lmm_pattern == LOV_PATTERN_MDT)
2442 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2444 i = comp_v1->lcm_entry_count > 1 ?
2445 comp_v1->lcm_entry_count - 1 : 0;
2446 ent = &comp_v1->lcm_entries[i];
2447 off = ent->lcme_offset;
2448 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2449 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2454 *lmm_size = lmmsize;
2459 static int ll_lov_setea(struct inode *inode, struct file *file,
2462 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2463 struct lov_user_md *lump;
2464 int lum_size = sizeof(struct lov_user_md) +
2465 sizeof(struct lov_user_ost_data);
2469 if (!capable(CAP_SYS_ADMIN))
2472 OBD_ALLOC_LARGE(lump, lum_size);
2476 if (copy_from_user(lump, arg, lum_size))
2477 GOTO(out_lump, rc = -EFAULT);
2479 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2481 cl_lov_delay_create_clear(&file->f_flags);
2484 OBD_FREE_LARGE(lump, lum_size);
2488 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2495 env = cl_env_get(&refcheck);
2497 RETURN(PTR_ERR(env));
2499 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2500 cl_env_put(env, &refcheck);
2504 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2507 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2508 struct lov_user_md *klum;
2510 __u64 flags = FMODE_WRITE;
2513 rc = ll_copy_user_md(lum, &klum);
2518 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2523 rc = put_user(0, &lum->lmm_stripe_count);
2527 rc = ll_layout_refresh(inode, &gen);
2531 rc = ll_file_getstripe(inode, arg, lum_size);
2532 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2533 ll_i2info(inode)->lli_clob) {
2534 struct iattr attr = { 0 };
2536 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2537 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2540 cl_lov_delay_create_clear(&file->f_flags);
2543 OBD_FREE_LARGE(klum, lum_size);
2549 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2551 struct ll_inode_info *lli = ll_i2info(inode);
2552 struct cl_object *obj = lli->lli_clob;
2553 struct ll_file_data *fd = file->private_data;
2554 struct ll_grouplock grouplock;
2559 CWARN("group id for group lock must not be 0\n");
2563 if (ll_file_nolock(file))
2564 RETURN(-EOPNOTSUPP);
2566 if (file->f_flags & O_NONBLOCK) {
2567 if (!mutex_trylock(&lli->lli_group_mutex))
2570 mutex_lock(&lli->lli_group_mutex);
2572 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2573 CWARN("group lock already existed with gid %lu\n",
2574 fd->fd_grouplock.lg_gid);
2575 GOTO(out, rc = -EINVAL);
2577 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2578 if (file->f_flags & O_NONBLOCK)
2579 GOTO(out, rc = -EAGAIN);
2580 mutex_unlock(&lli->lli_group_mutex);
2581 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2582 GOTO(retry, rc = 0);
2584 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2587 * XXX: group lock needs to protect all OST objects while PFL
2588 * can add new OST objects during the IO, so we'd instantiate
2589 * all OST objects before getting its group lock.
2594 struct cl_layout cl = {
2595 .cl_is_composite = false,
2597 struct lu_extent ext = {
2599 .e_end = OBD_OBJECT_EOF,
2602 env = cl_env_get(&refcheck);
2604 GOTO(out, rc = PTR_ERR(env));
2606 rc = cl_object_layout_get(env, obj, &cl);
2607 if (rc >= 0 && cl.cl_is_composite)
2608 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2611 cl_env_put(env, &refcheck);
2616 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2617 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2622 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2623 fd->fd_grouplock = grouplock;
2624 if (lli->lli_group_users == 0)
2625 lli->lli_group_gid = grouplock.lg_gid;
2626 lli->lli_group_users++;
2628 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2630 mutex_unlock(&lli->lli_group_mutex);
2635 static int ll_put_grouplock(struct inode *inode, struct file *file,
2638 struct ll_inode_info *lli = ll_i2info(inode);
2639 struct ll_file_data *fd = file->private_data;
2640 struct ll_grouplock grouplock;
2644 mutex_lock(&lli->lli_group_mutex);
2645 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2646 CWARN("no group lock held\n");
2647 GOTO(out, rc = -EINVAL);
2650 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2652 if (fd->fd_grouplock.lg_gid != arg) {
2653 CWARN("group lock %lu doesn't match current id %lu\n",
2654 arg, fd->fd_grouplock.lg_gid);
2655 GOTO(out, rc = -EINVAL);
2658 grouplock = fd->fd_grouplock;
2659 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2660 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2662 cl_put_grouplock(&grouplock);
2664 lli->lli_group_users--;
2665 if (lli->lli_group_users == 0) {
2666 lli->lli_group_gid = 0;
2667 wake_up_var(&lli->lli_group_users);
2669 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2672 mutex_unlock(&lli->lli_group_mutex);
2678 * Close inode open handle
2680 * \param dentry [in] dentry which contains the inode
2681 * \param it [in,out] intent which contains open info and result
2684 * \retval <0 failure
2686 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2688 struct inode *inode = dentry->d_inode;
2689 struct obd_client_handle *och;
2695 /* Root ? Do nothing. */
2696 if (is_root_inode(inode))
2699 /* No open handle to close? Move away */
2700 if (!it_disposition(it, DISP_OPEN_OPEN))
2703 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2705 OBD_ALLOC(och, sizeof(*och));
2707 GOTO(out, rc = -ENOMEM);
2709 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2713 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2715 /* this one is in place of ll_file_open */
2716 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2717 ptlrpc_req_finished(it->it_request);
2718 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2724 * Get size for inode for which FIEMAP mapping is requested.
2725 * Make the FIEMAP get_info call and returns the result.
2726 * \param fiemap kernel buffer to hold extens
2727 * \param num_bytes kernel buffer size
2729 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2735 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2738 /* Checks for fiemap flags */
2739 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2740 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2744 /* Check for FIEMAP_FLAG_SYNC */
2745 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2746 rc = filemap_fdatawrite(inode->i_mapping);
2751 env = cl_env_get(&refcheck);
2753 RETURN(PTR_ERR(env));
2755 if (i_size_read(inode) == 0) {
2756 rc = ll_glimpse_size(inode);
2761 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2762 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2763 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2765 /* If filesize is 0, then there would be no objects for mapping */
2766 if (fmkey.lfik_oa.o_size == 0) {
2767 fiemap->fm_mapped_extents = 0;
2771 fmkey.lfik_fiemap = *fiemap;
2773 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2774 &fmkey, fiemap, &num_bytes);
2776 cl_env_put(env, &refcheck);
2780 int ll_fid2path(struct inode *inode, void __user *arg)
2782 struct obd_export *exp = ll_i2mdexp(inode);
2783 const struct getinfo_fid2path __user *gfin = arg;
2785 struct getinfo_fid2path *gfout;
2791 if (!capable(CAP_DAC_READ_SEARCH) &&
2792 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
2795 /* Only need to get the buflen */
2796 if (get_user(pathlen, &gfin->gf_pathlen))
2799 if (pathlen > PATH_MAX)
2802 outsize = sizeof(*gfout) + pathlen;
2803 OBD_ALLOC(gfout, outsize);
2807 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2808 GOTO(gf_free, rc = -EFAULT);
2809 /* append root FID after gfout to let MDT know the root FID so that it
2810 * can lookup the correct path, this is mainly for fileset.
2811 * old server without fileset mount support will ignore this. */
2812 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2814 /* Call mdc_iocontrol */
2815 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2819 if (copy_to_user(arg, gfout, outsize))
2823 OBD_FREE(gfout, outsize);
2828 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2830 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2838 ioc->idv_version = 0;
2839 ioc->idv_layout_version = UINT_MAX;
2841 /* If no file object initialized, we consider its version is 0. */
2845 env = cl_env_get(&refcheck);
2847 RETURN(PTR_ERR(env));
2849 io = vvp_env_thread_io(env);
2851 io->u.ci_data_version.dv_data_version = 0;
2852 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2853 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2856 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2857 result = cl_io_loop(env, io);
2859 result = io->ci_result;
2861 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2862 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2864 cl_io_fini(env, io);
2866 if (unlikely(io->ci_need_restart))
2869 cl_env_put(env, &refcheck);
2875 * Read the data_version for inode.
2877 * This value is computed using stripe object version on OST.
2878 * Version is computed using server side locking.
2880 * @param flags if do sync on the OST side;
2882 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2883 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2885 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2887 struct ioc_data_version ioc = { .idv_flags = flags };
2890 rc = ll_ioc_data_version(inode, &ioc);
2892 *data_version = ioc.idv_version;
2898 * Trigger a HSM release request for the provided inode.
2900 int ll_hsm_release(struct inode *inode)
2903 struct obd_client_handle *och = NULL;
2904 __u64 data_version = 0;
2910 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2911 ll_i2sbi(inode)->ll_fsname,
2912 PFID(&ll_i2info(inode)->lli_fid));
2914 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2916 GOTO(out, rc = PTR_ERR(och));
2918 /* Grab latest data_version and [am]time values */
2919 rc = ll_data_version(inode, &data_version,
2920 LL_DV_WR_FLUSH | LL_DV_SZ_UPDATE);
2924 env = cl_env_get(&refcheck);
2926 GOTO(out, rc = PTR_ERR(env));
2928 rc = ll_merge_attr(env, inode);
2929 cl_env_put(env, &refcheck);
2931 /* If error happen, we have the wrong size for a file.
2937 /* Release the file.
2938 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2939 * we still need it to pack l_remote_handle to MDT. */
2940 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2946 if (och != NULL && !IS_ERR(och)) /* close the file */
2947 ll_lease_close(och, inode, NULL);
2952 struct ll_swap_stack {
2955 struct inode *inode1;
2956 struct inode *inode2;
2961 static int ll_swap_layouts(struct file *file1, struct file *file2,
2962 struct lustre_swap_layouts *lsl)
2964 struct mdc_swap_layouts msl;
2965 struct md_op_data *op_data;
2968 struct ll_swap_stack *llss = NULL;
2971 OBD_ALLOC_PTR(llss);
2975 llss->inode1 = file_inode(file1);
2976 llss->inode2 = file_inode(file2);
2978 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2982 /* we use 2 bool because it is easier to swap than 2 bits */
2983 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2984 llss->check_dv1 = true;
2986 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2987 llss->check_dv2 = true;
2989 /* we cannot use lsl->sl_dvX directly because we may swap them */
2990 llss->dv1 = lsl->sl_dv1;
2991 llss->dv2 = lsl->sl_dv2;
2993 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
2994 if (rc == 0) /* same file, done! */
2997 if (rc < 0) { /* sequentialize it */
2998 swap(llss->inode1, llss->inode2);
3000 swap(llss->dv1, llss->dv2);
3001 swap(llss->check_dv1, llss->check_dv2);
3005 if (gid != 0) { /* application asks to flush dirty cache */
3006 rc = ll_get_grouplock(llss->inode1, file1, gid);
3010 rc = ll_get_grouplock(llss->inode2, file2, gid);
3012 ll_put_grouplock(llss->inode1, file1, gid);
3017 /* ultimate check, before swaping the layouts we check if
3018 * dataversion has changed (if requested) */
3019 if (llss->check_dv1) {
3020 rc = ll_data_version(llss->inode1, &dv, 0);
3023 if (dv != llss->dv1)
3024 GOTO(putgl, rc = -EAGAIN);
3027 if (llss->check_dv2) {
3028 rc = ll_data_version(llss->inode2, &dv, 0);
3031 if (dv != llss->dv2)
3032 GOTO(putgl, rc = -EAGAIN);
3035 /* struct md_op_data is used to send the swap args to the mdt
3036 * only flags is missing, so we use struct mdc_swap_layouts
3037 * through the md_op_data->op_data */
3038 /* flags from user space have to be converted before they are send to
3039 * server, no flag is sent today, they are only used on the client */
3042 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3043 0, LUSTRE_OPC_ANY, &msl);
3044 if (IS_ERR(op_data))
3045 GOTO(free, rc = PTR_ERR(op_data));
3047 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3048 sizeof(*op_data), op_data, NULL);
3049 ll_finish_md_op_data(op_data);
3056 ll_put_grouplock(llss->inode2, file2, gid);
3057 ll_put_grouplock(llss->inode1, file1, gid);
3067 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3069 struct obd_export *exp = ll_i2mdexp(inode);
3070 struct md_op_data *op_data;
3074 /* Detect out-of range masks */
3075 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3078 /* Non-root users are forbidden to set or clear flags which are
3079 * NOT defined in HSM_USER_MASK. */
3080 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3081 !capable(CAP_SYS_ADMIN))
3084 if (!exp_connect_archive_id_array(exp)) {
3085 /* Detect out-of range archive id */
3086 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3087 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3091 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3092 LUSTRE_OPC_ANY, hss);
3093 if (IS_ERR(op_data))
3094 RETURN(PTR_ERR(op_data));
3096 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3099 ll_finish_md_op_data(op_data);
3104 static int ll_hsm_import(struct inode *inode, struct file *file,
3105 struct hsm_user_import *hui)
3107 struct hsm_state_set *hss = NULL;
3108 struct iattr *attr = NULL;
3112 if (!S_ISREG(inode->i_mode))
3118 GOTO(out, rc = -ENOMEM);
3120 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3121 hss->hss_archive_id = hui->hui_archive_id;
3122 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3123 rc = ll_hsm_state_set(inode, hss);
3127 OBD_ALLOC_PTR(attr);
3129 GOTO(out, rc = -ENOMEM);
3131 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3132 attr->ia_mode |= S_IFREG;
3133 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3134 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3135 attr->ia_size = hui->hui_size;
3136 attr->ia_mtime.tv_sec = hui->hui_mtime;
3137 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3138 attr->ia_atime.tv_sec = hui->hui_atime;
3139 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3141 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3142 ATTR_UID | ATTR_GID |
3143 ATTR_MTIME | ATTR_MTIME_SET |
3144 ATTR_ATIME | ATTR_ATIME_SET;
3148 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3152 inode_unlock(inode);
3164 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3166 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3167 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3170 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3172 struct inode *inode = file_inode(file);
3174 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3175 ATTR_MTIME | ATTR_MTIME_SET |
3178 .tv_sec = lfu->lfu_atime_sec,
3179 .tv_nsec = lfu->lfu_atime_nsec,
3182 .tv_sec = lfu->lfu_mtime_sec,
3183 .tv_nsec = lfu->lfu_mtime_nsec,
3186 .tv_sec = lfu->lfu_ctime_sec,
3187 .tv_nsec = lfu->lfu_ctime_nsec,
3193 if (!capable(CAP_SYS_ADMIN))
3196 if (!S_ISREG(inode->i_mode))
3200 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3202 inode_unlock(inode);
3207 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3210 case MODE_READ_USER:
3212 case MODE_WRITE_USER:
3219 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3221 /* Used to allow the upper layers of the client to request an LDLM lock
3222 * without doing an actual read or write.
3224 * Used for ladvise lockahead to manually request specific locks.
3226 * \param[in] file file this ladvise lock request is on
3227 * \param[in] ladvise ladvise struct describing this lock request
3229 * \retval 0 success, no detailed result available (sync requests
3230 * and requests sent to the server [not handled locally]
3231 * cannot return detailed results)
3232 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3233 * see definitions for details.
3234 * \retval negative negative errno on error
3236 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3238 struct lu_env *env = NULL;
3239 struct cl_io *io = NULL;
3240 struct cl_lock *lock = NULL;
3241 struct cl_lock_descr *descr = NULL;
3242 struct dentry *dentry = file->f_path.dentry;
3243 struct inode *inode = dentry->d_inode;
3244 enum cl_lock_mode cl_mode;
3245 off_t start = ladvise->lla_start;
3246 off_t end = ladvise->lla_end;
3253 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3254 dentry, dentry->d_inode,
3255 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3258 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3260 GOTO(out, result = cl_mode);
3262 /* Get IO environment */
3263 result = cl_io_get(inode, &env, &io, &refcheck);
3267 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3270 * nothing to do for this io. This currently happens when
3271 * stripe sub-object's are not yet created.
3273 result = io->ci_result;
3274 } else if (result == 0) {
3275 lock = vvp_env_lock(env);
3276 descr = &lock->cll_descr;
3278 descr->cld_obj = io->ci_obj;
3279 /* Convert byte offsets to pages */
3280 descr->cld_start = cl_index(io->ci_obj, start);
3281 descr->cld_end = cl_index(io->ci_obj, end);
3282 descr->cld_mode = cl_mode;
3283 /* CEF_MUST is used because we do not want to convert a
3284 * lockahead request to a lockless lock */
3285 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3287 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3288 descr->cld_enq_flags |= CEF_SPECULATIVE;
3290 result = cl_lock_request(env, io, lock);
3292 /* On success, we need to release the lock */
3294 cl_lock_release(env, lock);
3296 cl_io_fini(env, io);
3297 cl_env_put(env, &refcheck);
3299 /* -ECANCELED indicates a matching lock with a different extent
3300 * was already present, and -EEXIST indicates a matching lock
3301 * on exactly the same extent was already present.
3302 * We convert them to positive values for userspace to make
3303 * recognizing true errors easier.
3304 * Note we can only return these detailed results on async requests,
3305 * as sync requests look the same as i/o requests for locking. */
3306 if (result == -ECANCELED)
3307 result = LLA_RESULT_DIFFERENT;
3308 else if (result == -EEXIST)
3309 result = LLA_RESULT_SAME;
3314 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3316 static int ll_ladvise_sanity(struct inode *inode,
3317 struct llapi_lu_ladvise *ladvise)
3319 struct ll_sb_info *sbi = ll_i2sbi(inode);
3320 enum lu_ladvise_type advice = ladvise->lla_advice;
3321 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3322 * be in the first 32 bits of enum ladvise_flags */
3323 __u32 flags = ladvise->lla_peradvice_flags;
3324 /* 3 lines at 80 characters per line, should be plenty */
3327 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3330 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3331 sbi->ll_fsname, advice,
3332 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3336 /* Per-advice checks */
3338 case LU_LADVISE_LOCKNOEXPAND:
3339 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3341 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3342 "rc = %d\n", sbi->ll_fsname, flags,
3343 ladvise_names[advice], rc);
3347 case LU_LADVISE_LOCKAHEAD:
3348 /* Currently only READ and WRITE modes can be requested */
3349 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3350 ladvise->lla_lockahead_mode == 0) {
3352 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3353 "rc = %d\n", sbi->ll_fsname,
3354 ladvise->lla_lockahead_mode,
3355 ladvise_names[advice], rc);
3359 case LU_LADVISE_WILLREAD:
3360 case LU_LADVISE_DONTNEED:
3362 /* Note fall through above - These checks apply to all advices
3363 * except LOCKNOEXPAND */
3364 if (flags & ~LF_DEFAULT_MASK) {
3366 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3367 "rc = %d\n", sbi->ll_fsname, flags,
3368 ladvise_names[advice], rc);
3371 if (ladvise->lla_start >= ladvise->lla_end) {
3373 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3374 "for %s: rc = %d\n", sbi->ll_fsname,
3375 ladvise->lla_start, ladvise->lla_end,
3376 ladvise_names[advice], rc);
3388 * Give file access advices
3390 * The ladvise interface is similar to Linux fadvise() system call, except it
3391 * forwards the advices directly from Lustre client to server. The server side
3392 * codes will apply appropriate read-ahead and caching techniques for the
3393 * corresponding files.
3395 * A typical workload for ladvise is e.g. a bunch of different clients are
3396 * doing small random reads of a file, so prefetching pages into OSS cache
3397 * with big linear reads before the random IO is a net benefit. Fetching
3398 * all that data into each client cache with fadvise() may not be, due to
3399 * much more data being sent to the client.
3401 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3402 struct llapi_lu_ladvise *ladvise)
3406 struct cl_ladvise_io *lio;
3411 env = cl_env_get(&refcheck);
3413 RETURN(PTR_ERR(env));
3415 io = vvp_env_thread_io(env);
3416 io->ci_obj = ll_i2info(inode)->lli_clob;
3418 /* initialize parameters for ladvise */
3419 lio = &io->u.ci_ladvise;
3420 lio->li_start = ladvise->lla_start;
3421 lio->li_end = ladvise->lla_end;
3422 lio->li_fid = ll_inode2fid(inode);
3423 lio->li_advice = ladvise->lla_advice;
3424 lio->li_flags = flags;
3426 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3427 rc = cl_io_loop(env, io);
3431 cl_io_fini(env, io);
3432 cl_env_put(env, &refcheck);
3436 static int ll_lock_noexpand(struct file *file, int flags)
3438 struct ll_file_data *fd = file->private_data;
3440 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3445 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3448 struct fsxattr fsxattr;
3450 if (copy_from_user(&fsxattr,
3451 (const struct fsxattr __user *)arg,
3455 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3456 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3457 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3458 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3459 if (copy_to_user((struct fsxattr __user *)arg,
3460 &fsxattr, sizeof(fsxattr)))
3466 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3470 * Project Quota ID state is only allowed to change from within the init
3471 * namespace. Enforce that restriction only if we are trying to change
3472 * the quota ID state. Everything else is allowed in user namespaces.
3474 if (current_user_ns() == &init_user_ns) {
3476 * Caller is allowed to change the project ID. if it is being
3477 * changed, make sure that the new value is valid.
3479 if (ll_i2info(inode)->lli_projid != projid &&
3480 !projid_valid(make_kprojid(&init_user_ns, projid)))
3486 if (ll_i2info(inode)->lli_projid != projid)
3489 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3490 if (!(xflags & FS_XFLAG_PROJINHERIT))
3493 if (xflags & FS_XFLAG_PROJINHERIT)
3500 static int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3502 struct md_op_data *op_data;
3503 struct ptlrpc_request *req = NULL;
3504 struct cl_object *obj;
3505 unsigned int inode_flags;
3508 rc = ll_ioctl_check_project(inode, xflags, projid);
3512 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3513 LUSTRE_OPC_ANY, NULL);
3514 if (IS_ERR(op_data))
3515 RETURN(PTR_ERR(op_data));
3517 inode_flags = ll_xflags_to_inode_flags(xflags);
3518 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3519 if (xflags & FS_XFLAG_PROJINHERIT)
3520 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3521 op_data->op_projid = projid;
3522 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3523 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3524 ptlrpc_req_finished(req);
3526 GOTO(out_fsxattr, rc);
3527 ll_update_inode_flags(inode, op_data->op_attr_flags);
3529 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3530 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3531 GOTO(out_fsxattr, rc);
3533 obj = ll_i2info(inode)->lli_clob;
3535 struct iattr attr = { 0 };
3537 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3541 ll_finish_md_op_data(op_data);
3545 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3548 struct fsxattr fsxattr;
3552 if (copy_from_user(&fsxattr,
3553 (const struct fsxattr __user *)arg,
3557 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3558 fsxattr.fsx_projid));
3561 int ll_ioctl_project(struct file *file, unsigned int cmd,
3564 struct lu_project lu_project;
3565 struct dentry *dentry = file_dentry(file);
3566 struct inode *inode = file_inode(file);
3567 struct dentry *child_dentry = NULL;
3568 int rc = 0, name_len;
3570 if (copy_from_user(&lu_project,
3571 (const struct lu_project __user *)arg,
3572 sizeof(lu_project)))
3575 /* apply child dentry if name is valid */
3576 name_len = strnlen(lu_project.project_name, NAME_MAX);
3577 if (name_len > 0 && name_len <= NAME_MAX) {
3579 child_dentry = lookup_one_len(lu_project.project_name,
3581 inode_unlock(inode);
3582 if (IS_ERR(child_dentry)) {
3583 rc = PTR_ERR(child_dentry);
3586 inode = child_dentry->d_inode;
3591 } else if (name_len > NAME_MAX) {
3596 switch (lu_project.project_type) {
3597 case LU_PROJECT_SET:
3598 rc = ll_set_project(inode, lu_project.project_xflags,
3599 lu_project.project_id);
3601 case LU_PROJECT_GET:
3602 lu_project.project_xflags =
3603 ll_inode_flags_to_xflags(inode->i_flags);
3604 if (test_bit(LLIF_PROJECT_INHERIT,
3605 &ll_i2info(inode)->lli_flags))
3606 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3607 lu_project.project_id = ll_i2info(inode)->lli_projid;
3608 if (copy_to_user((struct lu_project __user *)arg,
3609 &lu_project, sizeof(lu_project))) {
3619 if (!IS_ERR_OR_NULL(child_dentry))
3624 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3627 struct inode *inode = file_inode(file);
3628 struct ll_file_data *fd = file->private_data;
3629 struct ll_inode_info *lli = ll_i2info(inode);
3630 struct obd_client_handle *och = NULL;
3631 struct split_param sp;
3632 struct pcc_param param;
3633 bool lease_broken = false;
3635 enum mds_op_bias bias = 0;
3636 struct file *layout_file = NULL;
3638 size_t data_size = 0;
3639 bool attached = false;
3644 mutex_lock(&lli->lli_och_mutex);
3645 if (fd->fd_lease_och != NULL) {
3646 och = fd->fd_lease_och;
3647 fd->fd_lease_och = NULL;
3649 mutex_unlock(&lli->lli_och_mutex);
3654 fmode = och->och_flags;
3656 switch (ioc->lil_flags) {
3657 case LL_LEASE_RESYNC_DONE:
3658 if (ioc->lil_count > IOC_IDS_MAX)
3659 GOTO(out_lease_close, rc = -EINVAL);
3661 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3662 OBD_ALLOC(data, data_size);
3664 GOTO(out_lease_close, rc = -ENOMEM);
3666 if (copy_from_user(data, (void __user *)arg, data_size))
3667 GOTO(out_lease_close, rc = -EFAULT);
3669 bias = MDS_CLOSE_RESYNC_DONE;
3671 case LL_LEASE_LAYOUT_MERGE: {
3674 if (ioc->lil_count != 1)
3675 GOTO(out_lease_close, rc = -EINVAL);
3677 arg += sizeof(*ioc);
3678 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3679 GOTO(out_lease_close, rc = -EFAULT);
3681 layout_file = fget(fd);
3683 GOTO(out_lease_close, rc = -EBADF);
3685 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3686 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3687 GOTO(out_lease_close, rc = -EPERM);
3689 data = file_inode(layout_file);
3690 bias = MDS_CLOSE_LAYOUT_MERGE;
3693 case LL_LEASE_LAYOUT_SPLIT: {
3697 if (ioc->lil_count != 2)
3698 GOTO(out_lease_close, rc = -EINVAL);
3700 arg += sizeof(*ioc);
3701 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3702 GOTO(out_lease_close, rc = -EFAULT);
3704 arg += sizeof(__u32);
3705 if (copy_from_user(&mirror_id, (void __user *)arg,
3707 GOTO(out_lease_close, rc = -EFAULT);
3709 layout_file = fget(fdv);
3711 GOTO(out_lease_close, rc = -EBADF);
3713 /* if layout_file == file, it means to destroy the mirror */
3714 sp.sp_inode = file_inode(layout_file);
3715 sp.sp_mirror_id = (__u16)mirror_id;
3717 bias = MDS_CLOSE_LAYOUT_SPLIT;
3720 case LL_LEASE_PCC_ATTACH:
3721 if (ioc->lil_count != 1)
3724 if (IS_ENCRYPTED(inode))
3725 RETURN(-EOPNOTSUPP);
3727 arg += sizeof(*ioc);
3728 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3730 GOTO(out_lease_close, rc2 = -EFAULT);
3732 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3734 GOTO(out_lease_close, rc2);
3737 /* Grab latest data version */
3738 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3741 GOTO(out_lease_close, rc2);
3744 bias = MDS_PCC_ATTACH;
3747 /* without close intent */
3752 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3756 rc = ll_lease_och_release(inode, file);
3765 switch (ioc->lil_flags) {
3766 case LL_LEASE_RESYNC_DONE:
3768 OBD_FREE(data, data_size);
3770 case LL_LEASE_LAYOUT_MERGE:
3771 case LL_LEASE_LAYOUT_SPLIT:
3775 ll_layout_refresh(inode, &fd->fd_layout_version);
3777 case LL_LEASE_PCC_ATTACH:
3780 rc = pcc_readwrite_attach_fini(file, inode,
3781 param.pa_layout_gen,
3788 rc = ll_lease_type_from_fmode(fmode);
3792 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3795 struct inode *inode = file_inode(file);
3796 struct ll_inode_info *lli = ll_i2info(inode);
3797 struct ll_file_data *fd = file->private_data;
3798 struct obd_client_handle *och = NULL;
3799 __u64 open_flags = 0;
3805 switch (ioc->lil_mode) {
3806 case LL_LEASE_WRLCK:
3807 if (!(file->f_mode & FMODE_WRITE))
3809 fmode = FMODE_WRITE;
3811 case LL_LEASE_RDLCK:
3812 if (!(file->f_mode & FMODE_READ))
3816 case LL_LEASE_UNLCK:
3817 RETURN(ll_file_unlock_lease(file, ioc, arg));
3822 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3824 /* apply for lease */
3825 if (ioc->lil_flags & LL_LEASE_RESYNC)
3826 open_flags = MDS_OPEN_RESYNC;
3827 och = ll_lease_open(inode, file, fmode, open_flags);
3829 RETURN(PTR_ERR(och));
3831 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3832 rc = ll_lease_file_resync(och, inode, arg);
3834 ll_lease_close(och, inode, NULL);
3837 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3839 ll_lease_close(och, inode, NULL);
3845 mutex_lock(&lli->lli_och_mutex);
3846 if (fd->fd_lease_och == NULL) {
3847 fd->fd_lease_och = och;
3850 mutex_unlock(&lli->lli_och_mutex);
3852 /* impossible now that only excl is supported for now */
3853 ll_lease_close(och, inode, &lease_broken);
3859 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3861 struct ll_inode_info *lli = ll_i2info(inode);
3862 struct ll_sb_info *sbi = ll_i2sbi(inode);
3863 __u64 now = ktime_get_real_seconds();
3866 spin_lock(&lli->lli_heat_lock);
3867 heat->lh_flags = lli->lli_heat_flags;
3868 for (i = 0; i < heat->lh_count; i++)
3869 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3870 now, sbi->ll_heat_decay_weight,
3871 sbi->ll_heat_period_second);
3872 spin_unlock(&lli->lli_heat_lock);
3875 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3877 struct ll_inode_info *lli = ll_i2info(inode);
3880 spin_lock(&lli->lli_heat_lock);
3881 if (flags & LU_HEAT_FLAG_CLEAR)
3882 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3884 if (flags & LU_HEAT_FLAG_OFF)
3885 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3887 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3889 spin_unlock(&lli->lli_heat_lock);
3895 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3897 struct inode *inode = file_inode(file);
3898 struct ll_file_data *fd = file->private_data;
3902 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3903 PFID(ll_inode2fid(inode)), inode, cmd);
3904 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3906 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3907 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3911 case LL_IOC_GETFLAGS:
3912 /* Get the current value of the file flags */
3913 return put_user(fd->fd_flags, (int __user *)arg);
3914 case LL_IOC_SETFLAGS:
3915 case LL_IOC_CLRFLAGS:
3916 /* Set or clear specific file flags */
3917 /* XXX This probably needs checks to ensure the flags are
3918 * not abused, and to handle any flag side effects.
3920 if (get_user(flags, (int __user *) arg))
3923 if (cmd == LL_IOC_SETFLAGS) {
3924 if ((flags & LL_FILE_IGNORE_LOCK) &&
3925 !(file->f_flags & O_DIRECT)) {
3926 CERROR("%s: unable to disable locking on "
3927 "non-O_DIRECT file\n", current->comm);
3931 fd->fd_flags |= flags;
3933 fd->fd_flags &= ~flags;
3936 case LL_IOC_LOV_SETSTRIPE:
3937 case LL_IOC_LOV_SETSTRIPE_NEW:
3938 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3939 case LL_IOC_LOV_SETEA:
3940 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3941 case LL_IOC_LOV_SWAP_LAYOUTS: {
3943 struct lustre_swap_layouts lsl;
3945 if (copy_from_user(&lsl, (char __user *)arg,
3946 sizeof(struct lustre_swap_layouts)))
3949 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3952 file2 = fget(lsl.sl_fd);
3956 /* O_WRONLY or O_RDWR */
3957 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3958 GOTO(out, rc = -EPERM);
3960 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3961 struct inode *inode2;
3962 struct ll_inode_info *lli;
3963 struct obd_client_handle *och = NULL;
3965 lli = ll_i2info(inode);
3966 mutex_lock(&lli->lli_och_mutex);
3967 if (fd->fd_lease_och != NULL) {
3968 och = fd->fd_lease_och;
3969 fd->fd_lease_och = NULL;
3971 mutex_unlock(&lli->lli_och_mutex);
3973 GOTO(out, rc = -ENOLCK);
3974 inode2 = file_inode(file2);
3975 rc = ll_swap_layouts_close(och, inode, inode2);
3977 rc = ll_swap_layouts(file, file2, &lsl);
3983 case LL_IOC_LOV_GETSTRIPE:
3984 case LL_IOC_LOV_GETSTRIPE_NEW:
3985 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
3986 case FS_IOC_GETFLAGS:
3987 case FS_IOC_SETFLAGS:
3988 RETURN(ll_iocontrol(inode, file, cmd, arg));
3989 case FSFILT_IOC_GETVERSION:
3990 case FS_IOC_GETVERSION:
3991 RETURN(put_user(inode->i_generation, (int __user *)arg));
3992 /* We need to special case any other ioctls we want to handle,
3993 * to send them to the MDS/OST as appropriate and to properly
3994 * network encode the arg field. */
3995 case FS_IOC_SETVERSION:
3998 case LL_IOC_GROUP_LOCK:
3999 RETURN(ll_get_grouplock(inode, file, arg));
4000 case LL_IOC_GROUP_UNLOCK:
4001 RETURN(ll_put_grouplock(inode, file, arg));
4002 case IOC_OBD_STATFS:
4003 RETURN(ll_obd_statfs(inode, (void __user *)arg));
4005 case LL_IOC_FLUSHCTX:
4006 RETURN(ll_flush_ctx(inode));
4007 case LL_IOC_PATH2FID: {
4008 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
4009 sizeof(struct lu_fid)))
4014 case LL_IOC_GETPARENT:
4015 RETURN(ll_getparent(file, (struct getparent __user *)arg));
4017 case OBD_IOC_FID2PATH:
4018 RETURN(ll_fid2path(inode, (void __user *)arg));
4019 case LL_IOC_DATA_VERSION: {
4020 struct ioc_data_version idv;
4023 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
4026 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
4027 rc = ll_ioc_data_version(inode, &idv);
4030 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
4036 case LL_IOC_GET_MDTIDX: {
4039 mdtidx = ll_get_mdt_idx(inode);
4043 if (put_user((int)mdtidx, (int __user *)arg))
4048 case OBD_IOC_GETNAME_OLD:
4049 case OBD_IOC_GETDTNAME:
4050 case OBD_IOC_GETMDNAME:
4051 RETURN(ll_get_obd_name(inode, cmd, arg));
4052 case LL_IOC_HSM_STATE_GET: {
4053 struct md_op_data *op_data;
4054 struct hsm_user_state *hus;
4061 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4062 LUSTRE_OPC_ANY, hus);
4063 if (IS_ERR(op_data)) {
4065 RETURN(PTR_ERR(op_data));
4068 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4071 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
4074 ll_finish_md_op_data(op_data);
4078 case LL_IOC_HSM_STATE_SET: {
4079 struct hsm_state_set *hss;
4086 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
4091 rc = ll_hsm_state_set(inode, hss);
4096 case LL_IOC_HSM_ACTION: {
4097 struct md_op_data *op_data;
4098 struct hsm_current_action *hca;
4106 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4107 LUSTRE_OPC_ANY, hca);
4108 if (IS_ERR(op_data)) {
4110 RETURN(PTR_ERR(op_data));
4113 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4116 GOTO(skip_copy, rc);
4118 /* The hsm_current_action retreived from the server could
4119 * contain corrupt information. If it is incorrect data collect
4120 * debug information. We still send the data even if incorrect
4121 * to user land to handle.
4123 action = hsm_user_action2name(hca->hca_action);
4124 if (strcmp(action, "UNKNOWN") == 0 ||
4125 hca->hca_state > HPS_DONE) {
4127 "HSM current state %s action %s, offset = %llu, length %llu\n",
4128 hsm_progress_state2name(hca->hca_state), action,
4129 hca->hca_location.offset, hca->hca_location.length);
4132 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
4135 ll_finish_md_op_data(op_data);
4139 case LL_IOC_SET_LEASE_OLD: {
4140 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
4142 RETURN(ll_file_set_lease(file, &ioc, 0));
4144 case LL_IOC_SET_LEASE: {
4145 struct ll_ioc_lease ioc;
4147 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
4150 RETURN(ll_file_set_lease(file, &ioc, arg));
4152 case LL_IOC_GET_LEASE: {
4153 struct ll_inode_info *lli = ll_i2info(inode);
4154 struct ldlm_lock *lock = NULL;
4157 mutex_lock(&lli->lli_och_mutex);
4158 if (fd->fd_lease_och != NULL) {
4159 struct obd_client_handle *och = fd->fd_lease_och;
4161 lock = ldlm_handle2lock(&och->och_lease_handle);
4163 lock_res_and_lock(lock);
4164 if (!ldlm_is_cancel(lock))
4165 fmode = och->och_flags;
4167 unlock_res_and_lock(lock);
4168 LDLM_LOCK_PUT(lock);
4171 mutex_unlock(&lli->lli_och_mutex);
4173 RETURN(ll_lease_type_from_fmode(fmode));
4175 case LL_IOC_HSM_IMPORT: {
4176 struct hsm_user_import *hui;
4182 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
4187 rc = ll_hsm_import(inode, file, hui);
4192 case LL_IOC_FUTIMES_3: {
4193 struct ll_futimes_3 lfu;
4195 if (copy_from_user(&lfu,
4196 (const struct ll_futimes_3 __user *)arg,
4200 RETURN(ll_file_futimes_3(file, &lfu));
4202 case LL_IOC_LADVISE: {
4203 struct llapi_ladvise_hdr *k_ladvise_hdr;
4204 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4207 int alloc_size = sizeof(*k_ladvise_hdr);
4210 u_ladvise_hdr = (void __user *)arg;
4211 OBD_ALLOC_PTR(k_ladvise_hdr);
4212 if (k_ladvise_hdr == NULL)
4215 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4216 GOTO(out_ladvise, rc = -EFAULT);
4218 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4219 k_ladvise_hdr->lah_count < 1)
4220 GOTO(out_ladvise, rc = -EINVAL);
4222 num_advise = k_ladvise_hdr->lah_count;
4223 if (num_advise >= LAH_COUNT_MAX)
4224 GOTO(out_ladvise, rc = -EFBIG);
4226 OBD_FREE_PTR(k_ladvise_hdr);
4227 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4228 lah_advise[num_advise]);
4229 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4230 if (k_ladvise_hdr == NULL)
4234 * TODO: submit multiple advices to one server in a single RPC
4236 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4237 GOTO(out_ladvise, rc = -EFAULT);
4239 for (i = 0; i < num_advise; i++) {
4240 struct llapi_lu_ladvise *k_ladvise =
4241 &k_ladvise_hdr->lah_advise[i];
4242 struct llapi_lu_ladvise __user *u_ladvise =
4243 &u_ladvise_hdr->lah_advise[i];
4245 rc = ll_ladvise_sanity(inode, k_ladvise);
4247 GOTO(out_ladvise, rc);
4249 switch (k_ladvise->lla_advice) {
4250 case LU_LADVISE_LOCKNOEXPAND:
4251 rc = ll_lock_noexpand(file,
4252 k_ladvise->lla_peradvice_flags);
4253 GOTO(out_ladvise, rc);
4254 case LU_LADVISE_LOCKAHEAD:
4256 rc = ll_file_lock_ahead(file, k_ladvise);
4259 GOTO(out_ladvise, rc);
4262 &u_ladvise->lla_lockahead_result))
4263 GOTO(out_ladvise, rc = -EFAULT);
4266 rc = ll_ladvise(inode, file,
4267 k_ladvise_hdr->lah_flags,
4270 GOTO(out_ladvise, rc);
4277 OBD_FREE(k_ladvise_hdr, alloc_size);
4280 case LL_IOC_FLR_SET_MIRROR: {
4281 /* mirror I/O must be direct to avoid polluting page cache
4283 if (!(file->f_flags & O_DIRECT))
4286 fd->fd_designated_mirror = (__u32)arg;
4289 case FS_IOC_FSGETXATTR:
4290 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
4291 case FS_IOC_FSSETXATTR:
4292 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
4293 case LL_IOC_PROJECT:
4294 RETURN(ll_ioctl_project(file, cmd, arg));
4296 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
4297 case LL_IOC_HEAT_GET: {
4298 struct lu_heat uheat;
4299 struct lu_heat *heat;
4302 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
4305 if (uheat.lh_count > OBD_HEAT_COUNT)
4306 uheat.lh_count = OBD_HEAT_COUNT;
4308 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4309 OBD_ALLOC(heat, size);
4313 heat->lh_count = uheat.lh_count;
4314 ll_heat_get(inode, heat);
4315 rc = copy_to_user((char __user *)arg, heat, size);
4316 OBD_FREE(heat, size);
4317 RETURN(rc ? -EFAULT : 0);
4319 case LL_IOC_HEAT_SET: {
4322 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
4325 rc = ll_heat_set(inode, flags);
4328 case LL_IOC_PCC_DETACH: {
4329 struct lu_pcc_detach *detach;
4331 OBD_ALLOC_PTR(detach);
4335 if (copy_from_user(detach,
4336 (const struct lu_pcc_detach __user *)arg,
4338 GOTO(out_detach_free, rc = -EFAULT);
4340 if (!S_ISREG(inode->i_mode))
4341 GOTO(out_detach_free, rc = -EINVAL);
4343 if (!inode_owner_or_capable(&init_user_ns, inode))
4344 GOTO(out_detach_free, rc = -EPERM);
4346 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4348 OBD_FREE_PTR(detach);
4351 case LL_IOC_PCC_STATE: {
4352 struct lu_pcc_state __user *ustate =
4353 (struct lu_pcc_state __user *)arg;
4354 struct lu_pcc_state *state;
4356 OBD_ALLOC_PTR(state);
4360 if (copy_from_user(state, ustate, sizeof(*state)))
4361 GOTO(out_state, rc = -EFAULT);
4363 rc = pcc_ioctl_state(file, inode, state);
4365 GOTO(out_state, rc);
4367 if (copy_to_user(ustate, state, sizeof(*state)))
4368 GOTO(out_state, rc = -EFAULT);
4371 OBD_FREE_PTR(state);
4374 #ifdef HAVE_LUSTRE_CRYPTO
4375 case LL_IOC_SET_ENCRYPTION_POLICY:
4376 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4378 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
4379 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
4380 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4382 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
4383 case LL_IOC_ADD_ENCRYPTION_KEY:
4384 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4386 return llcrypt_ioctl_add_key(file, (void __user *)arg);
4387 case LL_IOC_REMOVE_ENCRYPTION_KEY:
4388 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4390 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
4391 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
4392 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4394 return llcrypt_ioctl_remove_key_all_users(file,
4395 (void __user *)arg);
4396 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
4397 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4399 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
4402 case LL_IOC_UNLOCK_FOREIGN: {
4403 struct dentry *dentry = file_dentry(file);
4405 /* if not a foreign symlink do nothing */
4406 if (ll_foreign_is_removable(dentry, true)) {
4408 "prevent unlink of non-foreign file ("DFID")\n",
4409 PFID(ll_inode2fid(inode)));
4410 RETURN(-EOPNOTSUPP);
4416 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
4417 (void __user *)arg));
4421 loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4423 struct inode *inode = file_inode(file);
4426 struct cl_lseek_io *lsio;
4433 env = cl_env_get(&refcheck);
4435 RETURN(PTR_ERR(env));
4437 io = vvp_env_thread_io(env);
4438 io->ci_obj = ll_i2info(inode)->lli_clob;
4439 ll_io_set_mirror(io, file);
4441 lsio = &io->u.ci_lseek;
4442 lsio->ls_start = offset;
4443 lsio->ls_whence = whence;
4444 lsio->ls_result = -ENXIO;
4447 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4449 struct vvp_io *vio = vvp_env_io(env);
4451 vio->vui_fd = file->private_data;
4452 rc = cl_io_loop(env, io);
4456 retval = rc ? : lsio->ls_result;
4457 cl_io_fini(env, io);
4458 } while (unlikely(io->ci_need_restart));
4460 cl_env_put(env, &refcheck);
4462 /* Without the key, SEEK_HOLE return value has to be
4463 * rounded up to next LUSTRE_ENCRYPTION_UNIT_SIZE.
4465 if (llcrypt_require_key(inode) == -ENOKEY && whence == SEEK_HOLE)
4466 retval = round_up(retval, LUSTRE_ENCRYPTION_UNIT_SIZE);
4471 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4473 struct inode *inode = file_inode(file);
4474 loff_t retval = offset, eof = 0;
4475 ktime_t kstart = ktime_get();
4479 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4480 PFID(ll_inode2fid(inode)), inode, retval, retval,
4483 if (origin == SEEK_END) {
4484 retval = ll_glimpse_size(inode);
4487 eof = i_size_read(inode);
4490 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4494 /* flush local cache first if any */
4495 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4498 retval = ll_lseek(file, offset, origin);
4501 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4503 retval = generic_file_llseek_size(file, offset, origin,
4504 ll_file_maxbytes(inode), eof);
4507 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4508 ktime_us_delta(ktime_get(), kstart));
4512 static int ll_flush(struct file *file, fl_owner_t id)
4514 struct inode *inode = file_inode(file);
4515 struct ll_inode_info *lli = ll_i2info(inode);
4516 struct ll_file_data *fd = file->private_data;
4519 LASSERT(!S_ISDIR(inode->i_mode));
4521 /* catch async errors that were recorded back when async writeback
4522 * failed for pages in this mapping. */
4523 rc = lli->lli_async_rc;
4524 lli->lli_async_rc = 0;
4525 if (lli->lli_clob != NULL) {
4526 err = lov_read_and_clear_async_rc(lli->lli_clob);
4531 /* The application has been told write failure already.
4532 * Do not report failure again. */
4533 if (fd->fd_write_failed)
4535 return rc ? -EIO : 0;
4539 * Called to make sure a portion of file has been written out.
4540 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4542 * Return how many pages have been written.
4544 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4545 enum cl_fsync_mode mode, int ignore_layout)
4549 struct cl_fsync_io *fio;
4554 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4555 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4558 env = cl_env_get(&refcheck);
4560 RETURN(PTR_ERR(env));
4562 io = vvp_env_thread_io(env);
4563 io->ci_obj = ll_i2info(inode)->lli_clob;
4564 io->ci_ignore_layout = ignore_layout;
4566 /* initialize parameters for sync */
4567 fio = &io->u.ci_fsync;
4568 fio->fi_start = start;
4570 fio->fi_fid = ll_inode2fid(inode);
4571 fio->fi_mode = mode;
4572 fio->fi_nr_written = 0;
4574 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4575 result = cl_io_loop(env, io);
4577 result = io->ci_result;
4579 result = fio->fi_nr_written;
4580 cl_io_fini(env, io);
4581 cl_env_put(env, &refcheck);
4587 * When dentry is provided (the 'else' case), file_dentry() may be
4588 * null and dentry must be used directly rather than pulled from
4589 * file_dentry() as is done otherwise.
4592 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4594 struct dentry *dentry = file_dentry(file);
4595 struct inode *inode = dentry->d_inode;
4596 struct ll_inode_info *lli = ll_i2info(inode);
4597 struct ptlrpc_request *req;
4598 ktime_t kstart = ktime_get();
4604 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4605 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4607 /* fsync's caller has already called _fdata{sync,write}, we want
4608 * that IO to finish before calling the osc and mdc sync methods */
4609 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4611 /* catch async errors that were recorded back when async writeback
4612 * failed for pages in this mapping. */
4613 if (!S_ISDIR(inode->i_mode)) {
4614 err = lli->lli_async_rc;
4615 lli->lli_async_rc = 0;
4618 if (lli->lli_clob != NULL) {
4619 err = lov_read_and_clear_async_rc(lli->lli_clob);
4625 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4629 ptlrpc_req_finished(req);
4631 if (S_ISREG(inode->i_mode)) {
4632 struct ll_file_data *fd = file->private_data;
4635 /* Sync metadata on MDT first, and then sync the cached data
4638 err = pcc_fsync(file, start, end, datasync, &cached);
4640 err = cl_sync_file_range(inode, start, end,
4642 if (rc == 0 && err < 0)
4645 fd->fd_write_failed = true;
4647 fd->fd_write_failed = false;
4651 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4652 ktime_us_delta(ktime_get(), kstart));
4657 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4659 struct inode *inode = file_inode(file);
4660 struct ll_sb_info *sbi = ll_i2sbi(inode);
4661 struct ldlm_enqueue_info einfo = {
4662 .ei_type = LDLM_FLOCK,
4663 .ei_cb_cp = ldlm_flock_completion_ast,
4664 .ei_cbdata = file_lock,
4666 struct md_op_data *op_data;
4667 struct lustre_handle lockh = { 0 };
4668 union ldlm_policy_data flock = { { 0 } };
4669 int fl_type = file_lock->fl_type;
4670 ktime_t kstart = ktime_get();
4676 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4677 PFID(ll_inode2fid(inode)), file_lock);
4679 if (file_lock->fl_flags & FL_FLOCK) {
4680 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4681 /* flocks are whole-file locks */
4682 flock.l_flock.end = OFFSET_MAX;
4683 /* For flocks owner is determined by the local file desctiptor*/
4684 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4685 } else if (file_lock->fl_flags & FL_POSIX) {
4686 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4687 flock.l_flock.start = file_lock->fl_start;
4688 flock.l_flock.end = file_lock->fl_end;
4692 flock.l_flock.pid = file_lock->fl_pid;
4694 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4695 /* Somewhat ugly workaround for svc lockd.
4696 * lockd installs custom fl_lmops->lm_compare_owner that checks
4697 * for the fl_owner to be the same (which it always is on local node
4698 * I guess between lockd processes) and then compares pid.
4699 * As such we assign pid to the owner field to make it all work,
4700 * conflict with normal locks is unlikely since pid space and
4701 * pointer space for current->files are not intersecting */
4702 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4703 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4708 einfo.ei_mode = LCK_PR;
4711 /* An unlock request may or may not have any relation to
4712 * existing locks so we may not be able to pass a lock handle
4713 * via a normal ldlm_lock_cancel() request. The request may even
4714 * unlock a byte range in the middle of an existing lock. In
4715 * order to process an unlock request we need all of the same
4716 * information that is given with a normal read or write record
4717 * lock request. To avoid creating another ldlm unlock (cancel)
4718 * message we'll treat a LCK_NL flock request as an unlock. */
4719 einfo.ei_mode = LCK_NL;
4722 einfo.ei_mode = LCK_PW;
4725 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4740 flags = LDLM_FL_BLOCK_NOWAIT;
4746 flags = LDLM_FL_TEST_LOCK;
4749 CERROR("unknown fcntl lock command: %d\n", cmd);
4753 /* Save the old mode so that if the mode in the lock changes we
4754 * can decrement the appropriate reader or writer refcount. */
4755 file_lock->fl_type = einfo.ei_mode;
4757 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4758 LUSTRE_OPC_ANY, NULL);
4759 if (IS_ERR(op_data))
4760 RETURN(PTR_ERR(op_data));
4762 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4763 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4764 flock.l_flock.pid, flags, einfo.ei_mode,
4765 flock.l_flock.start, flock.l_flock.end);
4767 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4770 /* Restore the file lock type if not TEST lock. */
4771 if (!(flags & LDLM_FL_TEST_LOCK))
4772 file_lock->fl_type = fl_type;
4774 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4775 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4776 !(flags & LDLM_FL_TEST_LOCK))
4777 rc2 = locks_lock_file_wait(file, file_lock);
4779 if ((file_lock->fl_flags & FL_FLOCK) &&
4780 (rc == 0 || file_lock->fl_type == F_UNLCK))
4781 rc2 = flock_lock_file_wait(file, file_lock);
4782 if ((file_lock->fl_flags & FL_POSIX) &&
4783 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4784 !(flags & LDLM_FL_TEST_LOCK))
4785 rc2 = posix_lock_file_wait(file, file_lock);
4786 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4788 if (rc2 && file_lock->fl_type != F_UNLCK) {
4789 einfo.ei_mode = LCK_NL;
4790 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4795 ll_finish_md_op_data(op_data);
4798 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
4799 ktime_us_delta(ktime_get(), kstart));
4803 int ll_get_fid_by_name(struct inode *parent, const char *name,
4804 int namelen, struct lu_fid *fid,
4805 struct inode **inode)
4807 struct md_op_data *op_data = NULL;
4808 struct mdt_body *body;
4809 struct ptlrpc_request *req;
4813 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4814 LUSTRE_OPC_ANY, NULL);
4815 if (IS_ERR(op_data))
4816 RETURN(PTR_ERR(op_data));
4818 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4819 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4820 ll_finish_md_op_data(op_data);
4824 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4826 GOTO(out_req, rc = -EFAULT);
4828 *fid = body->mbo_fid1;
4831 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
4833 ptlrpc_req_finished(req);
4837 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4838 const char *name, __u32 flags)
4840 struct dentry *dchild = NULL;
4841 struct inode *child_inode = NULL;
4842 struct md_op_data *op_data;
4843 struct ptlrpc_request *request = NULL;
4844 struct obd_client_handle *och = NULL;
4846 struct mdt_body *body;
4847 __u64 data_version = 0;
4848 size_t namelen = strlen(name);
4849 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4853 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4854 PFID(ll_inode2fid(parent)), name,
4855 lum->lum_stripe_offset, lum->lum_stripe_count);
4857 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4858 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4859 lustre_swab_lmv_user_md(lum);
4861 /* Get child FID first */
4862 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4865 dchild = d_lookup(file_dentry(file), &qstr);
4867 if (dchild->d_inode)
4868 child_inode = igrab(dchild->d_inode);
4873 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4882 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4883 OBD_CONNECT2_DIR_MIGRATE)) {
4884 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4885 ll_dir_striped(child_inode)) {
4886 CERROR("%s: MDT doesn't support stripe directory "
4887 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4888 GOTO(out_iput, rc = -EOPNOTSUPP);
4893 * lfs migrate command needs to be blocked on the client
4894 * by checking the migrate FID against the FID of the
4897 if (is_root_inode(child_inode))
4898 GOTO(out_iput, rc = -EINVAL);
4900 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4901 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
4902 if (IS_ERR(op_data))
4903 GOTO(out_iput, rc = PTR_ERR(op_data));
4905 inode_lock(child_inode);
4906 op_data->op_fid3 = *ll_inode2fid(child_inode);
4907 if (!fid_is_sane(&op_data->op_fid3)) {
4908 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4909 ll_i2sbi(parent)->ll_fsname, name,
4910 PFID(&op_data->op_fid3));
4911 GOTO(out_unlock, rc = -EINVAL);
4914 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4915 op_data->op_data = lum;
4916 op_data->op_data_size = lumlen;
4918 /* migrate dirent only for subdirs if MDS_MIGRATE_NSONLY set */
4919 if (S_ISDIR(child_inode->i_mode) && (flags & MDS_MIGRATE_NSONLY) &&
4920 lmv_dir_layout_changing(ll_i2info(parent)->lli_lsm_md))
4921 op_data->op_bias |= MDS_MIGRATE_NSONLY;
4924 if (S_ISREG(child_inode->i_mode)) {
4925 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4929 GOTO(out_unlock, rc);
4932 rc = ll_data_version(child_inode, &data_version,
4935 GOTO(out_close, rc);
4937 op_data->op_open_handle = och->och_open_handle;
4938 op_data->op_data_version = data_version;
4939 op_data->op_lease_handle = och->och_lease_handle;
4940 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4942 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4943 och->och_mod->mod_open_req->rq_replay = 0;
4944 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4947 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
4948 op_data->op_name, op_data->op_namelen,
4949 op_data->op_name, op_data->op_namelen, &request);
4951 LASSERT(request != NULL);
4952 ll_update_times(request, parent);
4955 if (rc == 0 || rc == -EAGAIN) {
4956 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4957 LASSERT(body != NULL);
4959 /* If the server does release layout lock, then we cleanup
4960 * the client och here, otherwise release it in out_close: */
4961 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4962 obd_mod_put(och->och_mod);
4963 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4965 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4971 if (request != NULL) {
4972 ptlrpc_req_finished(request);
4976 /* Try again if the lease has cancelled. */
4977 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4982 ll_lease_close(och, child_inode, NULL);
4984 clear_nlink(child_inode);
4986 inode_unlock(child_inode);
4987 ll_finish_md_op_data(op_data);
4994 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
4996 struct ll_file_data *fd = file->private_data;
5000 * In order to avoid flood of warning messages, only print one message
5001 * for one file. And the entire message rate on the client is limited
5002 * by CDEBUG_LIMIT too.
5004 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
5005 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
5006 CDEBUG_LIMIT(D_CONSOLE,
5007 "flock disabled, mount with '-o [local]flock' to enable\r\n");
5013 * test if some locks matching bits and l_req_mode are acquired
5014 * - bits can be in different locks
5015 * - if found clear the common lock bits in *bits
5016 * - the bits not found, are kept in *bits
5018 * \param bits [IN] searched lock bits [IN]
5019 * \param l_req_mode [IN] searched lock mode
5020 * \retval boolean, true iff all bits are found
5022 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
5024 struct lustre_handle lockh;
5025 union ldlm_policy_data policy;
5026 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
5027 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5036 fid = &ll_i2info(inode)->lli_fid;
5037 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5038 ldlm_lockname[mode]);
5040 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5041 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5042 policy.l_inodebits.bits = *bits & BIT(i);
5043 if (policy.l_inodebits.bits == 0)
5046 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
5047 &policy, mode, &lockh)) {
5048 struct ldlm_lock *lock;
5050 lock = ldlm_handle2lock(&lockh);
5053 ~(lock->l_policy_data.l_inodebits.bits);
5054 LDLM_LOCK_PUT(lock);
5056 *bits &= ~policy.l_inodebits.bits;
5063 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5064 struct lustre_handle *lockh, __u64 flags,
5065 enum ldlm_mode mode)
5067 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5072 fid = &ll_i2info(inode)->lli_fid;
5073 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5075 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5076 fid, LDLM_IBITS, &policy, mode, lockh);
5081 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5083 /* Already unlinked. Just update nlink and return success */
5084 if (rc == -ENOENT) {
5086 /* If it is striped directory, and there is bad stripe
5087 * Let's revalidate the dentry again, instead of returning
5089 if (ll_dir_striped(inode))
5092 /* This path cannot be hit for regular files unless in
5093 * case of obscure races, so no need to to validate
5095 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5097 } else if (rc != 0) {
5098 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5099 "%s: revalidate FID "DFID" error: rc = %d\n",
5100 ll_i2sbi(inode)->ll_fsname,
5101 PFID(ll_inode2fid(inode)), rc);
5107 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5109 struct inode *parent;
5110 struct inode *inode = dentry->d_inode;
5111 struct obd_export *exp = ll_i2mdexp(inode);
5112 struct lookup_intent oit = {
5115 struct ptlrpc_request *req = NULL;
5116 struct md_op_data *op_data;
5117 const char *name = NULL;
5122 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5123 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5125 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5126 parent = dentry->d_parent->d_inode;
5127 name = dentry->d_name.name;
5128 namelen = dentry->d_name.len;
5133 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5134 LUSTRE_OPC_ANY, NULL);
5135 if (IS_ERR(op_data))
5136 RETURN(PTR_ERR(op_data));
5138 /* Call getattr by fid */
5139 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5140 op_data->op_flags = MF_GETATTR_BY_FID;
5141 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5142 ll_finish_md_op_data(op_data);
5144 rc = ll_inode_revalidate_fini(inode, rc);
5148 rc = ll_revalidate_it_finish(req, &oit, dentry);
5150 ll_intent_release(&oit);
5154 /* Unlinked? Unhash dentry, so it is not picked up later by
5155 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5156 * here to preserve get_cwd functionality on 2.6.
5158 if (!dentry->d_inode->i_nlink)
5159 d_lustre_invalidate(dentry);
5161 ll_lookup_finish_locks(&oit, dentry);
5163 ptlrpc_req_finished(req);
5168 static int ll_merge_md_attr(struct inode *inode)
5170 struct ll_inode_info *lli = ll_i2info(inode);
5171 struct cl_attr attr = { 0 };
5174 LASSERT(lli->lli_lsm_md != NULL);
5176 if (!lmv_dir_striped(lli->lli_lsm_md))
5179 down_read(&lli->lli_lsm_sem);
5180 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
5181 &attr, ll_md_blocking_ast);
5182 up_read(&lli->lli_lsm_sem);
5186 spin_lock(&inode->i_lock);
5187 set_nlink(inode, attr.cat_nlink);
5188 spin_unlock(&inode->i_lock);
5190 inode->i_blocks = attr.cat_blocks;
5191 i_size_write(inode, attr.cat_size);
5193 ll_i2info(inode)->lli_atime = attr.cat_atime;
5194 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5195 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5200 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5201 unsigned int flags, bool foreign)
5203 struct inode *inode = de->d_inode;
5204 struct ll_sb_info *sbi = ll_i2sbi(inode);
5205 struct ll_inode_info *lli = ll_i2info(inode);
5206 struct inode *dir = de->d_parent->d_inode;
5207 bool need_glimpse = true;
5208 ktime_t kstart = ktime_get();
5211 /* The OST object(s) determine the file size, blocks and mtime. */
5212 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5213 request_mask & STATX_MTIME))
5214 need_glimpse = false;
5216 if (dentry_may_statahead(dir, de))
5217 ll_start_statahead(dir, de, need_glimpse &&
5218 !(flags & AT_STATX_DONT_SYNC));
5220 if (flags & AT_STATX_DONT_SYNC)
5221 GOTO(fill_attr, rc = 0);
5223 rc = ll_inode_revalidate(de, IT_GETATTR);
5227 /* foreign file/dir are always of zero length, so don't
5228 * need to validate size.
5230 if (S_ISREG(inode->i_mode) && !foreign) {
5234 GOTO(fill_attr, rc);
5236 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5237 if (cached && rc < 0)
5241 GOTO(fill_attr, rc);
5244 * If the returned attr is masked with OBD_MD_FLSIZE &
5245 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5246 * or blocks obtained from MDT is strictly correct, and the file
5247 * is usually not being modified by clients, and the [a|m|c]time
5248 * got from MDT is also strictly correct.
5249 * Under this circumstance, it does not need to send glimpse
5250 * RPCs to OSTs for file attributes such as the size and blocks.
5252 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5253 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5254 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5255 inode->i_mtime.tv_sec = lli->lli_mtime;
5256 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5257 inode->i_atime.tv_sec = lli->lli_atime;
5258 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5259 inode->i_ctime.tv_sec = lli->lli_ctime;
5260 GOTO(fill_attr, rc);
5263 /* In case of restore, the MDT has the right size and has
5264 * already send it back without granting the layout lock,
5265 * inode is up-to-date so glimpse is useless.
5266 * Also to glimpse we need the layout, in case of a running
5267 * restore the MDT holds the layout lock so the glimpse will
5268 * block up to the end of restore (getattr will block)
5270 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5271 rc = ll_glimpse_size(inode);
5276 /* If object isn't regular a file then don't validate size. */
5277 /* foreign dir is not striped dir */
5278 if (ll_dir_striped(inode) && !foreign) {
5279 rc = ll_merge_md_attr(inode);
5284 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5285 inode->i_atime.tv_sec = lli->lli_atime;
5286 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5287 inode->i_mtime.tv_sec = lli->lli_mtime;
5288 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5289 inode->i_ctime.tv_sec = lli->lli_ctime;
5293 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5295 if (ll_need_32bit_api(sbi)) {
5296 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5297 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5298 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5300 stat->ino = inode->i_ino;
5301 stat->dev = inode->i_sb->s_dev;
5302 stat->rdev = inode->i_rdev;
5305 /* foreign symlink to be exposed as a real symlink */
5307 stat->mode = inode->i_mode;
5309 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5311 stat->uid = inode->i_uid;
5312 stat->gid = inode->i_gid;
5313 stat->atime = inode->i_atime;
5314 stat->mtime = inode->i_mtime;
5315 stat->ctime = inode->i_ctime;
5316 /* stat->blksize is used to tell about preferred IO size */
5317 if (sbi->ll_stat_blksize)
5318 stat->blksize = sbi->ll_stat_blksize;
5319 else if (S_ISREG(inode->i_mode))
5320 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
5321 LL_MAX_BLKSIZE_BITS);
5323 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5325 stat->nlink = inode->i_nlink;
5326 stat->size = i_size_read(inode);
5327 stat->blocks = inode->i_blocks;
5329 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5330 if (flags & AT_STATX_DONT_SYNC) {
5331 if (stat->size == 0 &&
5332 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5333 stat->size = lli->lli_lazysize;
5334 if (stat->blocks == 0 &&
5335 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5336 stat->blocks = lli->lli_lazyblocks;
5339 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5340 stat->result_mask |= STATX_BTIME;
5341 stat->btime.tv_sec = lli->lli_btime;
5344 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5345 #ifdef HAVE_LUSTRE_CRYPTO
5346 stat->attributes_mask |= STATX_ATTR_ENCRYPTED;
5348 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5349 stat->result_mask &= request_mask;
5352 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5353 ktime_us_delta(ktime_get(), kstart));
5358 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5359 int ll_getattr(struct user_namespace *mnt_userns, const struct path *path,
5360 struct kstat *stat, u32 request_mask, unsigned int flags)
5362 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5366 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5368 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5369 AT_STATX_SYNC_AS_STAT, false);
5373 int cl_falloc(struct file *file, struct inode *inode, int mode, loff_t offset,
5380 loff_t size = i_size_read(inode);
5384 env = cl_env_get(&refcheck);
5386 RETURN(PTR_ERR(env));
5388 io = vvp_env_thread_io(env);
5389 io->ci_obj = ll_i2info(inode)->lli_clob;
5390 ll_io_set_mirror(io, file);
5392 io->ci_verify_layout = 1;
5393 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5394 io->u.ci_setattr.sa_falloc_mode = mode;
5395 io->u.ci_setattr.sa_falloc_offset = offset;
5396 io->u.ci_setattr.sa_falloc_end = offset + len;
5397 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5399 CDEBUG(D_INODE, "UID %u GID %u\n",
5400 from_kuid(&init_user_ns, inode->i_uid),
5401 from_kgid(&init_user_ns, inode->i_gid));
5403 io->u.ci_setattr.sa_falloc_uid = from_kuid(&init_user_ns, inode->i_uid);
5404 io->u.ci_setattr.sa_falloc_gid = from_kgid(&init_user_ns, inode->i_gid);
5406 if (io->u.ci_setattr.sa_falloc_end > size) {
5407 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5409 /* Check new size against VFS/VM file size limit and rlimit */
5410 rc = inode_newsize_ok(inode, newsize);
5413 if (newsize > ll_file_maxbytes(inode)) {
5414 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5415 (unsigned long long)newsize,
5416 ll_file_maxbytes(inode));
5423 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5425 rc = cl_io_loop(env, io);
5428 cl_io_fini(env, io);
5429 } while (unlikely(io->ci_need_restart));
5432 cl_env_put(env, &refcheck);
5436 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5438 struct inode *inode = file_inode(filp);
5441 if (offset < 0 || len <= 0)
5444 * Encrypted inodes can't handle collapse range or zero range or insert
5445 * range since we would need to re-encrypt blocks with a different IV or
5446 * XTS tweak (which are based on the logical block number).
5447 * Similar to what ext4 does.
5449 if (IS_ENCRYPTED(inode) &&
5450 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5451 FALLOC_FL_ZERO_RANGE)))
5452 RETURN(-EOPNOTSUPP);
5455 * mode == 0 (which is standard prealloc) and PUNCH is supported
5456 * Rest of mode options are not supported yet.
5458 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5459 RETURN(-EOPNOTSUPP);
5461 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5463 rc = cl_falloc(filp, inode, mode, offset, len);
5465 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5466 * used by Lustre in several places. Retuning it here would
5467 * confuse applications that explicity test for EOPNOTSUPP
5468 * (95) and fall back to ftruncate().
5470 if (rc == -ENOTSUPP)
5476 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5477 __u64 start, __u64 len)
5481 struct fiemap *fiemap;
5482 unsigned int extent_count = fieinfo->fi_extents_max;
5484 num_bytes = sizeof(*fiemap) + (extent_count *
5485 sizeof(struct fiemap_extent));
5486 OBD_ALLOC_LARGE(fiemap, num_bytes);
5491 fiemap->fm_flags = fieinfo->fi_flags;
5492 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5493 fiemap->fm_start = start;
5494 fiemap->fm_length = len;
5495 if (extent_count > 0 &&
5496 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5497 sizeof(struct fiemap_extent)) != 0)
5498 GOTO(out, rc = -EFAULT);
5500 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5502 if (IS_ENCRYPTED(inode)) {
5505 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5506 fiemap->fm_extents[i].fe_flags |=
5507 FIEMAP_EXTENT_DATA_ENCRYPTED |
5508 FIEMAP_EXTENT_ENCODED;
5511 fieinfo->fi_flags = fiemap->fm_flags;
5512 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5513 if (extent_count > 0 &&
5514 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5515 fiemap->fm_mapped_extents *
5516 sizeof(struct fiemap_extent)) != 0)
5517 GOTO(out, rc = -EFAULT);
5519 OBD_FREE_LARGE(fiemap, num_bytes);
5523 int ll_inode_permission(struct user_namespace *mnt_userns, struct inode *inode,
5527 struct ll_sb_info *sbi;
5528 struct root_squash_info *squash;
5529 struct cred *cred = NULL;
5530 const struct cred *old_cred = NULL;
5531 bool squash_id = false;
5532 ktime_t kstart = ktime_get();
5536 if (mask & MAY_NOT_BLOCK)
5540 * as root inode are NOT getting validated in lookup operation,
5541 * need to do it before permission check.
5544 if (is_root_inode(inode)) {
5545 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
5550 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5551 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5553 /* squash fsuid/fsgid if needed */
5554 sbi = ll_i2sbi(inode);
5555 squash = &sbi->ll_squash;
5556 if (unlikely(squash->rsi_uid != 0 &&
5557 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5558 !test_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags))) {
5562 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5563 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5564 squash->rsi_uid, squash->rsi_gid);
5566 /* update current process's credentials
5567 * and FS capability */
5568 cred = prepare_creds();
5572 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5573 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5574 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5575 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5577 old_cred = override_creds(cred);
5580 rc = generic_permission(mnt_userns, inode, mask);
5581 /* restore current process's credentials and FS capability */
5583 revert_creds(old_cred);
5588 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5589 ktime_us_delta(ktime_get(), kstart));
5594 /* -o localflock - only provides locally consistent flock locks */
5595 static const struct file_operations ll_file_operations = {
5596 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5597 # ifdef HAVE_SYNC_READ_WRITE
5598 .read = new_sync_read,
5599 .write = new_sync_write,
5601 .read_iter = ll_file_read_iter,
5602 .write_iter = ll_file_write_iter,
5603 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5604 .read = ll_file_read,
5605 .aio_read = ll_file_aio_read,
5606 .write = ll_file_write,
5607 .aio_write = ll_file_aio_write,
5608 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5609 .unlocked_ioctl = ll_file_ioctl,
5610 .open = ll_file_open,
5611 .release = ll_file_release,
5612 .mmap = ll_file_mmap,
5613 .llseek = ll_file_seek,
5614 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5615 .splice_read = generic_file_splice_read,
5617 .splice_read = pcc_file_splice_read,
5619 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5620 .splice_write = iter_file_splice_write,
5624 .fallocate = ll_fallocate,
5627 static const struct file_operations ll_file_operations_flock = {
5628 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5629 # ifdef HAVE_SYNC_READ_WRITE
5630 .read = new_sync_read,
5631 .write = new_sync_write,
5632 # endif /* HAVE_SYNC_READ_WRITE */
5633 .read_iter = ll_file_read_iter,
5634 .write_iter = ll_file_write_iter,
5635 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5636 .read = ll_file_read,
5637 .aio_read = ll_file_aio_read,
5638 .write = ll_file_write,
5639 .aio_write = ll_file_aio_write,
5640 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5641 .unlocked_ioctl = ll_file_ioctl,
5642 .open = ll_file_open,
5643 .release = ll_file_release,
5644 .mmap = ll_file_mmap,
5645 .llseek = ll_file_seek,
5646 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5647 .splice_read = generic_file_splice_read,
5649 .splice_read = pcc_file_splice_read,
5651 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5652 .splice_write = iter_file_splice_write,
5656 .flock = ll_file_flock,
5657 .lock = ll_file_flock,
5658 .fallocate = ll_fallocate,
5661 /* These are for -o noflock - to return ENOSYS on flock calls */
5662 static const struct file_operations ll_file_operations_noflock = {
5663 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5664 # ifdef HAVE_SYNC_READ_WRITE
5665 .read = new_sync_read,
5666 .write = new_sync_write,
5667 # endif /* HAVE_SYNC_READ_WRITE */
5668 .read_iter = ll_file_read_iter,
5669 .write_iter = ll_file_write_iter,
5670 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5671 .read = ll_file_read,
5672 .aio_read = ll_file_aio_read,
5673 .write = ll_file_write,
5674 .aio_write = ll_file_aio_write,
5675 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5676 .unlocked_ioctl = ll_file_ioctl,
5677 .open = ll_file_open,
5678 .release = ll_file_release,
5679 .mmap = ll_file_mmap,
5680 .llseek = ll_file_seek,
5681 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5682 .splice_read = generic_file_splice_read,
5684 .splice_read = pcc_file_splice_read,
5686 #ifdef HAVE_ITER_FILE_SPLICE_WRITE
5687 .splice_write = iter_file_splice_write,
5691 .flock = ll_file_noflock,
5692 .lock = ll_file_noflock,
5693 .fallocate = ll_fallocate,
5696 const struct inode_operations ll_file_inode_operations = {
5697 .setattr = ll_setattr,
5698 .getattr = ll_getattr,
5699 .permission = ll_inode_permission,
5700 #ifdef HAVE_IOP_XATTR
5701 .setxattr = ll_setxattr,
5702 .getxattr = ll_getxattr,
5703 .removexattr = ll_removexattr,
5705 .listxattr = ll_listxattr,
5706 .fiemap = ll_fiemap,
5707 .get_acl = ll_get_acl,
5708 #ifdef HAVE_IOP_SET_ACL
5709 .set_acl = ll_set_acl,
5713 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
5715 const struct file_operations *fops = &ll_file_operations_noflock;
5717 if (test_bit(LL_SBI_FLOCK, sbi->ll_flags))
5718 fops = &ll_file_operations_flock;
5719 else if (test_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags))
5720 fops = &ll_file_operations;
5725 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5727 struct ll_inode_info *lli = ll_i2info(inode);
5728 struct cl_object *obj = lli->lli_clob;
5737 env = cl_env_get(&refcheck);
5739 RETURN(PTR_ERR(env));
5741 rc = cl_conf_set(env, lli->lli_clob, conf);
5745 if (conf->coc_opc == OBJECT_CONF_SET) {
5746 struct ldlm_lock *lock = conf->coc_lock;
5747 struct cl_layout cl = {
5751 LASSERT(lock != NULL);
5752 LASSERT(ldlm_has_layout(lock));
5754 /* it can only be allowed to match after layout is
5755 * applied to inode otherwise false layout would be
5756 * seen. Applying layout shoud happen before dropping
5757 * the intent lock. */
5758 ldlm_lock_allow_match(lock);
5760 rc = cl_object_layout_get(env, obj, &cl);
5765 DFID": layout version change: %u -> %u\n",
5766 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5768 ll_layout_version_set(lli, cl.cl_layout_gen);
5772 cl_env_put(env, &refcheck);
5774 RETURN(rc < 0 ? rc : 0);
5777 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5778 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5781 struct ll_sb_info *sbi = ll_i2sbi(inode);
5782 struct ptlrpc_request *req;
5789 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5790 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5791 lock->l_lvb_data, lock->l_lvb_len);
5793 if (lock->l_lvb_data != NULL)
5796 /* if layout lock was granted right away, the layout is returned
5797 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5798 * blocked and then granted via completion ast, we have to fetch
5799 * layout here. Please note that we can't use the LVB buffer in
5800 * completion AST because it doesn't have a large enough buffer */
5801 rc = ll_get_default_mdsize(sbi, &lmmsize);
5805 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5806 XATTR_NAME_LOV, lmmsize, &req);
5809 GOTO(out, rc = 0); /* empty layout */
5816 if (lmmsize == 0) /* empty layout */
5819 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5821 GOTO(out, rc = -EFAULT);
5823 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5824 if (lvbdata == NULL)
5825 GOTO(out, rc = -ENOMEM);
5827 memcpy(lvbdata, lmm, lmmsize);
5828 lock_res_and_lock(lock);
5829 if (unlikely(lock->l_lvb_data == NULL)) {
5830 lock->l_lvb_type = LVB_T_LAYOUT;
5831 lock->l_lvb_data = lvbdata;
5832 lock->l_lvb_len = lmmsize;
5835 unlock_res_and_lock(lock);
5838 OBD_FREE_LARGE(lvbdata, lmmsize);
5843 ptlrpc_req_finished(req);
5848 * Apply the layout to the inode. Layout lock is held and will be released
5851 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5852 struct inode *inode)
5854 struct ll_inode_info *lli = ll_i2info(inode);
5855 struct ll_sb_info *sbi = ll_i2sbi(inode);
5856 struct ldlm_lock *lock;
5857 struct cl_object_conf conf;
5860 bool wait_layout = false;
5863 LASSERT(lustre_handle_is_used(lockh));
5865 lock = ldlm_handle2lock(lockh);
5866 LASSERT(lock != NULL);
5868 if (!ldlm_has_layout(lock))
5869 GOTO(out, rc = -EAGAIN);
5871 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5872 PFID(&lli->lli_fid), inode);
5874 /* in case this is a caching lock and reinstate with new inode */
5875 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5877 lock_res_and_lock(lock);
5878 lvb_ready = ldlm_is_lvb_ready(lock);
5879 unlock_res_and_lock(lock);
5881 /* checking lvb_ready is racy but this is okay. The worst case is
5882 * that multi processes may configure the file on the same time. */
5886 rc = ll_layout_fetch(inode, lock);
5890 /* for layout lock, lmm is stored in lock's lvb.
5891 * lvb_data is immutable if the lock is held so it's safe to access it
5894 * set layout to file. Unlikely this will fail as old layout was
5895 * surely eliminated */
5896 memset(&conf, 0, sizeof conf);
5897 conf.coc_opc = OBJECT_CONF_SET;
5898 conf.coc_inode = inode;
5899 conf.coc_lock = lock;
5900 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5901 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5902 rc = ll_layout_conf(inode, &conf);
5904 /* refresh layout failed, need to wait */
5905 wait_layout = rc == -EBUSY;
5908 LDLM_LOCK_PUT(lock);
5909 ldlm_lock_decref(lockh, mode);
5911 /* wait for IO to complete if it's still being used. */
5913 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5914 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5916 memset(&conf, 0, sizeof conf);
5917 conf.coc_opc = OBJECT_CONF_WAIT;
5918 conf.coc_inode = inode;
5919 rc = ll_layout_conf(inode, &conf);
5923 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5924 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5930 * Issue layout intent RPC to MDS.
5931 * \param inode [in] file inode
5932 * \param intent [in] layout intent
5934 * \retval 0 on success
5935 * \retval < 0 error code
5937 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5939 struct ll_inode_info *lli = ll_i2info(inode);
5940 struct ll_sb_info *sbi = ll_i2sbi(inode);
5941 struct md_op_data *op_data;
5942 struct lookup_intent it;
5943 struct ptlrpc_request *req;
5947 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5948 0, 0, LUSTRE_OPC_ANY, NULL);
5949 if (IS_ERR(op_data))
5950 RETURN(PTR_ERR(op_data));
5952 op_data->op_data = intent;
5953 op_data->op_data_size = sizeof(*intent);
5955 memset(&it, 0, sizeof(it));
5956 it.it_op = IT_LAYOUT;
5957 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5958 intent->li_opc == LAYOUT_INTENT_TRUNC)
5959 it.it_flags = FMODE_WRITE;
5961 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5962 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5964 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5965 &ll_md_blocking_ast, 0);
5966 if (it.it_request != NULL)
5967 ptlrpc_req_finished(it.it_request);
5968 it.it_request = NULL;
5970 ll_finish_md_op_data(op_data);
5972 /* set lock data in case this is a new lock */
5974 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5976 ll_intent_drop_lock(&it);
5982 * This function checks if there exists a LAYOUT lock on the client side,
5983 * or enqueues it if it doesn't have one in cache.
5985 * This function will not hold layout lock so it may be revoked any time after
5986 * this function returns. Any operations depend on layout should be redone
5989 * This function should be called before lov_io_init() to get an uptodate
5990 * layout version, the caller should save the version number and after IO
5991 * is finished, this function should be called again to verify that layout
5992 * is not changed during IO time.
5994 int ll_layout_refresh(struct inode *inode, __u32 *gen)
5996 struct ll_inode_info *lli = ll_i2info(inode);
5997 struct ll_sb_info *sbi = ll_i2sbi(inode);
5998 struct lustre_handle lockh;
5999 struct layout_intent intent = {
6000 .li_opc = LAYOUT_INTENT_ACCESS,
6002 enum ldlm_mode mode;
6006 *gen = ll_layout_version_get(lli);
6007 if (!test_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags) ||
6008 *gen != CL_LAYOUT_GEN_NONE)
6012 LASSERT(fid_is_sane(ll_inode2fid(inode)));
6013 LASSERT(S_ISREG(inode->i_mode));
6015 /* take layout lock mutex to enqueue layout lock exclusively. */
6016 mutex_lock(&lli->lli_layout_mutex);
6019 /* mostly layout lock is caching on the local side, so try to
6020 * match it before grabbing layout lock mutex. */
6021 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
6022 LCK_CR | LCK_CW | LCK_PR |
6024 if (mode != 0) { /* hit cached lock */
6025 rc = ll_layout_lock_set(&lockh, mode, inode);
6031 rc = ll_layout_intent(inode, &intent);
6037 *gen = ll_layout_version_get(lli);
6038 mutex_unlock(&lli->lli_layout_mutex);
6044 * Issue layout intent RPC indicating where in a file an IO is about to write.
6046 * \param[in] inode file inode.
6047 * \param[in] ext write range with start offset of fille in bytes where
6048 * an IO is about to write, and exclusive end offset in
6051 * \retval 0 on success
6052 * \retval < 0 error code
6054 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
6055 struct lu_extent *ext)
6057 struct layout_intent intent = {
6059 .li_extent.e_start = ext->e_start,
6060 .li_extent.e_end = ext->e_end,
6065 rc = ll_layout_intent(inode, &intent);
6071 * This function send a restore request to the MDT
6073 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6075 struct hsm_user_request *hur;
6079 len = sizeof(struct hsm_user_request) +
6080 sizeof(struct hsm_user_item);
6081 OBD_ALLOC(hur, len);
6085 hur->hur_request.hr_action = HUA_RESTORE;
6086 hur->hur_request.hr_archive_id = 0;
6087 hur->hur_request.hr_flags = 0;
6088 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6089 sizeof(hur->hur_user_item[0].hui_fid));
6090 hur->hur_user_item[0].hui_extent.offset = offset;
6091 hur->hur_user_item[0].hui_extent.length = length;
6092 hur->hur_request.hr_itemcount = 1;
6093 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,