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 <uapi/linux/llcrypt.h>
50 #include <lustre_swab.h>
52 #include "cl_object.h"
53 #include "llite_internal.h"
54 #include "vvp_internal.h"
57 struct inode *sp_inode;
62 __u64 pa_data_version;
68 ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg);
70 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
73 static struct ll_file_data *ll_file_data_get(void)
75 struct ll_file_data *fd;
77 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, GFP_NOFS);
81 fd->fd_write_failed = false;
82 pcc_file_init(&fd->fd_pcc_file);
87 static void ll_file_data_put(struct ll_file_data *fd)
90 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
94 * Packs all the attributes into @op_data for the CLOSE rpc.
96 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
97 struct obd_client_handle *och)
101 ll_prep_md_op_data(op_data, inode, NULL, NULL,
102 0, 0, LUSTRE_OPC_ANY, NULL);
104 op_data->op_attr.ia_mode = inode->i_mode;
105 op_data->op_attr.ia_atime = inode->i_atime;
106 op_data->op_attr.ia_mtime = inode->i_mtime;
107 op_data->op_attr.ia_ctime = inode->i_ctime;
108 /* In case of encrypted file without the key, visible size was rounded
109 * up to next LUSTRE_ENCRYPTION_UNIT_SIZE, and clear text size was
110 * stored into lli_lazysize in ll_merge_attr(), so set proper file size
111 * now that we are closing.
113 if (llcrypt_require_key(inode) == -ENOKEY &&
114 ll_i2info(inode)->lli_attr_valid & OBD_MD_FLLAZYSIZE)
115 op_data->op_attr.ia_size = ll_i2info(inode)->lli_lazysize;
117 op_data->op_attr.ia_size = i_size_read(inode);
118 op_data->op_attr.ia_valid |= (ATTR_MODE | ATTR_ATIME | ATTR_ATIME_SET |
119 ATTR_MTIME | ATTR_MTIME_SET |
121 op_data->op_xvalid |= OP_XVALID_CTIME_SET;
122 op_data->op_attr_blocks = inode->i_blocks;
123 op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
124 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
125 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
126 op_data->op_open_handle = och->och_open_handle;
128 if (och->och_flags & FMODE_WRITE &&
129 test_and_clear_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags))
130 /* For HSM: if inode data has been modified, pack it so that
131 * MDT can set data dirty flag in the archive. */
132 op_data->op_bias |= MDS_DATA_MODIFIED;
138 * Perform a close, possibly with a bias.
139 * The meaning of "data" depends on the value of "bias".
141 * If \a bias is MDS_HSM_RELEASE then \a data is a pointer to the data version.
142 * If \a bias is MDS_CLOSE_LAYOUT_SWAP then \a data is a pointer to the inode to
145 static int ll_close_inode_openhandle(struct inode *inode,
146 struct obd_client_handle *och,
147 enum mds_op_bias bias, void *data)
149 struct obd_export *md_exp = ll_i2mdexp(inode);
150 const struct ll_inode_info *lli = ll_i2info(inode);
151 struct md_op_data *op_data;
152 struct ptlrpc_request *req = NULL;
156 if (class_exp2obd(md_exp) == NULL) {
157 CERROR("%s: invalid MDC connection handle closing "DFID"\n",
158 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
162 OBD_ALLOC_PTR(op_data);
163 /* We leak openhandle and request here on error, but not much to be
164 * done in OOM case since app won't retry close on error either. */
166 GOTO(out, rc = -ENOMEM);
168 ll_prepare_close(inode, op_data, och);
170 case MDS_CLOSE_LAYOUT_MERGE:
171 /* merge blocks from the victim inode */
172 op_data->op_attr_blocks += ((struct inode *)data)->i_blocks;
173 op_data->op_attr.ia_valid |= ATTR_SIZE;
174 op_data->op_xvalid |= OP_XVALID_BLOCKS;
176 case MDS_CLOSE_LAYOUT_SPLIT:
177 case MDS_CLOSE_LAYOUT_SWAP: {
178 struct split_param *sp = data;
180 LASSERT(data != NULL);
181 op_data->op_bias |= bias;
182 op_data->op_data_version = 0;
183 op_data->op_lease_handle = och->och_lease_handle;
184 if (bias == MDS_CLOSE_LAYOUT_SPLIT) {
185 op_data->op_fid2 = *ll_inode2fid(sp->sp_inode);
186 op_data->op_mirror_id = sp->sp_mirror_id;
188 op_data->op_fid2 = *ll_inode2fid(data);
193 case MDS_CLOSE_RESYNC_DONE: {
194 struct ll_ioc_lease *ioc = data;
196 LASSERT(data != NULL);
197 op_data->op_attr_blocks +=
198 ioc->lil_count * op_data->op_attr_blocks;
199 op_data->op_attr.ia_valid |= ATTR_SIZE;
200 op_data->op_xvalid |= OP_XVALID_BLOCKS;
201 op_data->op_bias |= MDS_CLOSE_RESYNC_DONE;
203 op_data->op_lease_handle = och->och_lease_handle;
204 op_data->op_data = &ioc->lil_ids[0];
205 op_data->op_data_size =
206 ioc->lil_count * sizeof(ioc->lil_ids[0]);
210 case MDS_PCC_ATTACH: {
211 struct pcc_param *param = data;
213 LASSERT(data != NULL);
214 op_data->op_bias |= MDS_HSM_RELEASE | MDS_PCC_ATTACH;
215 op_data->op_archive_id = param->pa_archive_id;
216 op_data->op_data_version = param->pa_data_version;
217 op_data->op_lease_handle = och->och_lease_handle;
221 case MDS_HSM_RELEASE:
222 LASSERT(data != NULL);
223 op_data->op_bias |= MDS_HSM_RELEASE;
224 op_data->op_data_version = *(__u64 *)data;
225 op_data->op_lease_handle = och->och_lease_handle;
226 op_data->op_attr.ia_valid |= ATTR_SIZE;
227 op_data->op_xvalid |= OP_XVALID_BLOCKS;
231 LASSERT(data == NULL);
235 if (!(op_data->op_attr.ia_valid & ATTR_SIZE))
236 op_data->op_xvalid |= OP_XVALID_LAZYSIZE;
237 if (!(op_data->op_xvalid & OP_XVALID_BLOCKS))
238 op_data->op_xvalid |= OP_XVALID_LAZYBLOCKS;
240 rc = md_close(md_exp, op_data, och->och_mod, &req);
241 if (rc != 0 && rc != -EINTR)
242 CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
243 md_exp->exp_obd->obd_name, PFID(&lli->lli_fid), rc);
245 if (rc == 0 && op_data->op_bias & bias) {
246 struct mdt_body *body;
248 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
249 if (!(body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED))
252 if (bias & MDS_PCC_ATTACH) {
253 struct pcc_param *param = data;
255 param->pa_layout_gen = body->mbo_layout_gen;
259 ll_finish_md_op_data(op_data);
263 md_clear_open_replay_data(md_exp, och);
264 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
267 ptlrpc_req_finished(req); /* This is close request */
271 int ll_md_real_close(struct inode *inode, fmode_t fmode)
273 struct ll_inode_info *lli = ll_i2info(inode);
274 struct obd_client_handle **och_p;
275 struct obd_client_handle *och;
280 if (fmode & FMODE_WRITE) {
281 och_p = &lli->lli_mds_write_och;
282 och_usecount = &lli->lli_open_fd_write_count;
283 } else if (fmode & FMODE_EXEC) {
284 och_p = &lli->lli_mds_exec_och;
285 och_usecount = &lli->lli_open_fd_exec_count;
287 LASSERT(fmode & FMODE_READ);
288 och_p = &lli->lli_mds_read_och;
289 och_usecount = &lli->lli_open_fd_read_count;
292 mutex_lock(&lli->lli_och_mutex);
293 if (*och_usecount > 0) {
294 /* There are still users of this handle, so skip
296 mutex_unlock(&lli->lli_och_mutex);
302 mutex_unlock(&lli->lli_och_mutex);
305 /* There might be a race and this handle may already
307 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
313 static int ll_md_close(struct inode *inode, struct file *file)
315 union ldlm_policy_data policy = {
316 .l_inodebits = { MDS_INODELOCK_OPEN },
318 __u64 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
319 struct ll_file_data *fd = file->private_data;
320 struct ll_inode_info *lli = ll_i2info(inode);
321 struct lustre_handle lockh;
322 enum ldlm_mode lockmode;
326 /* clear group lock, if present */
327 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
328 ll_put_grouplock(inode, file, fd->fd_grouplock.lg_gid);
330 mutex_lock(&lli->lli_och_mutex);
331 if (fd->fd_lease_och != NULL) {
333 struct obd_client_handle *lease_och;
335 lease_och = fd->fd_lease_och;
336 fd->fd_lease_och = NULL;
337 mutex_unlock(&lli->lli_och_mutex);
339 /* Usually the lease is not released when the
340 * application crashed, we need to release here. */
341 rc = ll_lease_close(lease_och, inode, &lease_broken);
343 mutex_lock(&lli->lli_och_mutex);
345 CDEBUG_LIMIT(rc ? D_ERROR : D_INODE,
346 "Clean up lease "DFID" %d/%d\n",
347 PFID(&lli->lli_fid), rc, lease_broken);
350 if (fd->fd_och != NULL) {
351 struct obd_client_handle *och;
355 mutex_unlock(&lli->lli_och_mutex);
357 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
361 /* Let's see if we have good enough OPEN lock on the file and if
362 we can skip talking to MDS */
363 if (fd->fd_omode & FMODE_WRITE) {
365 LASSERT(lli->lli_open_fd_write_count);
366 lli->lli_open_fd_write_count--;
367 } else if (fd->fd_omode & FMODE_EXEC) {
369 LASSERT(lli->lli_open_fd_exec_count);
370 lli->lli_open_fd_exec_count--;
373 LASSERT(lli->lli_open_fd_read_count);
374 lli->lli_open_fd_read_count--;
376 mutex_unlock(&lli->lli_och_mutex);
378 /* LU-4398: do not cache write open lock if the file has exec bit */
379 if ((lockmode == LCK_CW && inode->i_mode & S_IXUGO) ||
380 !md_lock_match(ll_i2mdexp(inode), flags, ll_inode2fid(inode),
381 LDLM_IBITS, &policy, lockmode, &lockh))
382 rc = ll_md_real_close(inode, fd->fd_omode);
385 file->private_data = NULL;
386 ll_file_data_put(fd);
391 /* While this returns an error code, fput() the caller does not, so we need
392 * to make every effort to clean up all of our state here. Also, applications
393 * rarely check close errors and even if an error is returned they will not
394 * re-try the close call.
396 int ll_file_release(struct inode *inode, struct file *file)
398 struct ll_file_data *fd;
399 struct ll_sb_info *sbi = ll_i2sbi(inode);
400 struct ll_inode_info *lli = ll_i2info(inode);
401 ktime_t kstart = ktime_get();
406 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
407 PFID(ll_inode2fid(inode)), inode);
409 fd = file->private_data;
412 /* The last ref on @file, maybe not the the owner pid of statahead,
413 * because parent and child process can share the same file handle. */
414 if (S_ISDIR(inode->i_mode) && lli->lli_opendir_key == fd)
415 ll_deauthorize_statahead(inode, fd);
417 if (is_root_inode(inode)) {
418 file->private_data = NULL;
419 ll_file_data_put(fd);
423 pcc_file_release(inode, file);
425 if (!S_ISDIR(inode->i_mode)) {
426 if (lli->lli_clob != NULL)
427 lov_read_and_clear_async_rc(lli->lli_clob);
428 lli->lli_async_rc = 0;
431 lli->lli_close_fd_time = ktime_get();
433 rc = ll_md_close(inode, file);
435 if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val))
436 libcfs_debug_dumplog();
439 if (!rc && !is_root_inode(inode))
440 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE,
441 ktime_us_delta(ktime_get(), kstart));
445 static inline int ll_dom_readpage(void *data, struct page *page)
447 /* since ll_dom_readpage is a page cache helper, it is safe to assume
448 * mapping and host pointers are set here
451 struct niobuf_local *lnb = data;
455 inode = page2inode(page);
457 kaddr = kmap_atomic(page);
458 memcpy(kaddr, lnb->lnb_data, lnb->lnb_len);
459 if (lnb->lnb_len < PAGE_SIZE)
460 memset(kaddr + lnb->lnb_len, 0,
461 PAGE_SIZE - lnb->lnb_len);
462 flush_dcache_page(page);
463 SetPageUptodate(page);
464 kunmap_atomic(kaddr);
466 if (inode && IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
467 if (!llcrypt_has_encryption_key(inode))
468 CDEBUG(D_SEC, "no enc key for "DFID"\n",
469 PFID(ll_inode2fid(inode)));
471 unsigned int offs = 0;
473 while (offs < PAGE_SIZE) {
474 /* decrypt only if page is not empty */
475 if (memcmp(page_address(page) + offs,
476 page_address(ZERO_PAGE(0)),
477 LUSTRE_ENCRYPTION_UNIT_SIZE) == 0)
480 rc = llcrypt_decrypt_pagecache_blocks(page,
481 LUSTRE_ENCRYPTION_UNIT_SIZE,
486 offs += LUSTRE_ENCRYPTION_UNIT_SIZE;
495 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req)
499 struct ll_inode_info *lli = ll_i2info(inode);
500 struct cl_object *obj = lli->lli_clob;
501 struct address_space *mapping = inode->i_mapping;
503 struct niobuf_remote *rnb;
504 struct mdt_body *body;
506 unsigned long index, start;
507 struct niobuf_local lnb;
516 if (!req_capsule_field_present(&req->rq_pill, &RMF_NIOBUF_INLINE,
520 rnb = req_capsule_server_get(&req->rq_pill, &RMF_NIOBUF_INLINE);
521 if (rnb == NULL || rnb->rnb_len == 0)
524 /* LU-11595: Server may return whole file and that is OK always or
525 * it may return just file tail and its offset must be aligned with
526 * client PAGE_SIZE to be used on that client, if server's PAGE_SIZE is
527 * smaller then offset may be not aligned and that data is just ignored.
529 if (rnb->rnb_offset & ~PAGE_MASK)
532 /* Server returns whole file or just file tail if it fills in reply
533 * buffer, in both cases total size should be equal to the file size.
535 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
536 if (rnb->rnb_offset + rnb->rnb_len != body->mbo_dom_size &&
537 !(inode && IS_ENCRYPTED(inode))) {
538 CERROR("%s: server returns off/len %llu/%u but size %llu\n",
539 ll_i2sbi(inode)->ll_fsname, rnb->rnb_offset,
540 rnb->rnb_len, body->mbo_dom_size);
544 env = cl_env_get(&refcheck);
547 io = vvp_env_thread_io(env);
549 io->ci_ignore_layout = 1;
550 rc = cl_io_init(env, io, CIT_MISC, obj);
554 CDEBUG(D_INFO, "Get data along with open at %llu len %i, size %llu\n",
555 rnb->rnb_offset, rnb->rnb_len, body->mbo_dom_size);
557 data = (char *)rnb + sizeof(*rnb);
559 lnb.lnb_file_offset = rnb->rnb_offset;
560 start = lnb.lnb_file_offset >> PAGE_SHIFT;
562 LASSERT((lnb.lnb_file_offset & ~PAGE_MASK) == 0);
563 lnb.lnb_page_offset = 0;
565 struct cl_page *page;
567 lnb.lnb_data = data + (index << PAGE_SHIFT);
568 lnb.lnb_len = rnb->rnb_len - (index << PAGE_SHIFT);
569 if (lnb.lnb_len > PAGE_SIZE)
570 lnb.lnb_len = PAGE_SIZE;
572 vmpage = read_cache_page(mapping, index + start,
573 ll_dom_readpage, &lnb);
574 if (IS_ERR(vmpage)) {
575 CWARN("%s: cannot fill page %lu for "DFID
576 " with data: rc = %li\n",
577 ll_i2sbi(inode)->ll_fsname, index + start,
578 PFID(lu_object_fid(&obj->co_lu)),
583 if (vmpage->mapping == NULL) {
586 /* page was truncated */
589 /* attach VM page to CL page cache */
590 page = cl_page_find(env, obj, vmpage->index, vmpage,
593 ClearPageUptodate(vmpage);
598 cl_page_export(env, page, 1);
599 cl_page_put(env, page);
603 } while (rnb->rnb_len > (index << PAGE_SHIFT));
607 cl_env_put(env, &refcheck);
612 static int ll_intent_file_open(struct dentry *de, void *lmm, int lmmsize,
613 struct lookup_intent *itp)
615 struct ll_sb_info *sbi = ll_i2sbi(de->d_inode);
616 struct dentry *parent = de->d_parent;
619 struct md_op_data *op_data;
620 struct ptlrpc_request *req = NULL;
624 LASSERT(parent != NULL);
625 LASSERT(itp->it_flags & MDS_OPEN_BY_FID);
627 /* if server supports open-by-fid, or file name is invalid, don't pack
628 * name in open request */
629 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_OPEN_BY_NAME) ||
630 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_OPEN_BY_FID)) {
632 len = de->d_name.len;
633 name = kmalloc(len + 1, GFP_NOFS);
638 spin_lock(&de->d_lock);
639 if (len != de->d_name.len) {
640 spin_unlock(&de->d_lock);
644 memcpy(name, de->d_name.name, len);
646 spin_unlock(&de->d_lock);
648 if (!lu_name_is_valid_2(name, len)) {
654 op_data = ll_prep_md_op_data(NULL, parent->d_inode, de->d_inode,
655 name, len, 0, LUSTRE_OPC_OPEN, NULL);
656 if (IS_ERR(op_data)) {
658 RETURN(PTR_ERR(op_data));
660 op_data->op_data = lmm;
661 op_data->op_data_size = lmmsize;
663 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_OPEN_DELAY, cfs_fail_val);
665 rc = md_intent_lock(sbi->ll_md_exp, op_data, itp, &req,
666 &ll_md_blocking_ast, 0);
668 ll_finish_md_op_data(op_data);
670 /* reason for keep own exit path - don`t flood log
671 * with messages with -ESTALE errors.
673 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
674 it_open_error(DISP_OPEN_OPEN, itp))
676 ll_release_openhandle(de, itp);
680 if (it_disposition(itp, DISP_LOOKUP_NEG))
681 GOTO(out, rc = -ENOENT);
683 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
684 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
685 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
689 rc = ll_prep_inode(&de->d_inode, &req->rq_pill, NULL, itp);
691 if (!rc && itp->it_lock_mode) {
694 /* If we got a lock back and it has a LOOKUP bit set,
695 * make sure the dentry is marked as valid so we can find it.
696 * We don't need to care about actual hashing since other bits
697 * of kernel will deal with that later.
699 ll_set_lock_data(sbi->ll_md_exp, de->d_inode, itp, &bits);
700 if (bits & MDS_INODELOCK_LOOKUP) {
701 d_lustre_revalidate(de);
702 ll_update_dir_depth(parent->d_inode, de->d_inode);
705 /* if DoM bit returned along with LAYOUT bit then there
706 * can be read-on-open data returned.
708 if (bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
709 ll_dom_finish_open(de->d_inode, req);
713 ptlrpc_req_finished(req);
714 ll_intent_drop_lock(itp);
716 /* We did open by fid, but by the time we got to the server, the object
717 * disappeared. This is possible if the object was unlinked, but it's
718 * also possible if the object was unlinked by a rename. In the case
719 * of an object renamed over our existing one, we can't fail this open.
720 * O_CREAT also goes through this path if we had an existing dentry,
721 * and it's obviously wrong to return ENOENT for O_CREAT.
723 * Instead let's return -ESTALE, and the VFS will retry the open with
724 * LOOKUP_REVAL, which we catch in ll_revalidate_dentry and fail to
725 * revalidate, causing a lookup. This causes extra lookups in the case
726 * where we had a dentry in cache but the file is being unlinked and we
727 * lose the race with unlink, but this should be very rare.
735 static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
736 struct obd_client_handle *och)
738 struct mdt_body *body;
740 body = req_capsule_server_get(&it->it_request->rq_pill, &RMF_MDT_BODY);
741 och->och_open_handle = body->mbo_open_handle;
742 och->och_fid = body->mbo_fid1;
743 och->och_lease_handle.cookie = it->it_lock_handle;
744 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
745 och->och_flags = it->it_flags;
747 return md_set_open_replay_data(md_exp, och, it);
750 static int ll_local_open(struct file *file, struct lookup_intent *it,
751 struct ll_file_data *fd, struct obd_client_handle *och)
753 struct inode *inode = file_inode(file);
756 LASSERT(!file->private_data);
763 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
768 file->private_data = fd;
769 ll_readahead_init(inode, &fd->fd_ras);
770 fd->fd_omode = it->it_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
775 void ll_track_file_opens(struct inode *inode)
777 struct ll_inode_info *lli = ll_i2info(inode);
778 struct ll_sb_info *sbi = ll_i2sbi(inode);
780 /* do not skew results with delays from never-opened inodes */
781 if (ktime_to_ns(lli->lli_close_fd_time))
782 ll_stats_ops_tally(sbi, LPROC_LL_INODE_OPCLTM,
783 ktime_us_delta(ktime_get(), lli->lli_close_fd_time));
785 if (ktime_after(ktime_get(),
786 ktime_add_ms(lli->lli_close_fd_time,
787 sbi->ll_oc_max_ms))) {
788 lli->lli_open_fd_count = 1;
789 lli->lli_close_fd_time = ns_to_ktime(0);
791 lli->lli_open_fd_count++;
794 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_OCOUNT,
795 lli->lli_open_fd_count);
798 /* Open a file, and (for the very first open) create objects on the OSTs at
799 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
800 * creation or open until ll_lov_setstripe() ioctl is called.
802 * If we already have the stripe MD locally then we don't request it in
803 * md_open(), by passing a lmm_size = 0.
805 * It is up to the application to ensure no other processes open this file
806 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
807 * used. We might be able to avoid races of that sort by getting lli_open_sem
808 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
809 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
811 int ll_file_open(struct inode *inode, struct file *file)
813 struct ll_inode_info *lli = ll_i2info(inode);
814 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
815 .it_flags = file->f_flags };
816 struct obd_client_handle **och_p = NULL;
817 __u64 *och_usecount = NULL;
818 struct ll_file_data *fd;
819 ktime_t kstart = ktime_get();
823 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), flags %o\n",
824 PFID(ll_inode2fid(inode)), inode, file->f_flags);
826 it = file->private_data; /* XXX: compat macro */
827 file->private_data = NULL; /* prevent ll_local_open assertion */
829 if (S_ISREG(inode->i_mode)) {
830 rc = ll_file_open_encrypt(inode, file);
832 if (it && it->it_disposition)
833 ll_release_openhandle(file_dentry(file), it);
834 GOTO(out_nofiledata, rc);
838 fd = ll_file_data_get();
840 GOTO(out_nofiledata, rc = -ENOMEM);
843 if (S_ISDIR(inode->i_mode))
844 ll_authorize_statahead(inode, fd);
846 ll_track_file_opens(inode);
847 if (is_root_inode(inode)) {
848 file->private_data = fd;
852 if (!it || !it->it_disposition) {
853 /* Convert f_flags into access mode. We cannot use file->f_mode,
854 * because everything but O_ACCMODE mask was stripped from
856 if ((oit.it_flags + 1) & O_ACCMODE)
858 if (file->f_flags & O_TRUNC)
859 oit.it_flags |= FMODE_WRITE;
861 /* kernel only call f_op->open in dentry_open. filp_open calls
862 * dentry_open after call to open_namei that checks permissions.
863 * Only nfsd_open call dentry_open directly without checking
864 * permissions and because of that this code below is safe.
866 if (oit.it_flags & (FMODE_WRITE | FMODE_READ))
867 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
869 /* We do not want O_EXCL here, presumably we opened the file
870 * already? XXX - NFS implications? */
871 oit.it_flags &= ~O_EXCL;
873 /* bug20584, if "it_flags" contains O_CREAT, the file will be
874 * created if necessary, then "IT_CREAT" should be set to keep
875 * consistent with it */
876 if (oit.it_flags & O_CREAT)
877 oit.it_op |= IT_CREAT;
883 /* Let's see if we have file open on MDS already. */
884 if (it->it_flags & FMODE_WRITE) {
885 och_p = &lli->lli_mds_write_och;
886 och_usecount = &lli->lli_open_fd_write_count;
887 } else if (it->it_flags & FMODE_EXEC) {
888 och_p = &lli->lli_mds_exec_och;
889 och_usecount = &lli->lli_open_fd_exec_count;
891 och_p = &lli->lli_mds_read_och;
892 och_usecount = &lli->lli_open_fd_read_count;
895 mutex_lock(&lli->lli_och_mutex);
896 if (*och_p) { /* Open handle is present */
897 if (it_disposition(it, DISP_OPEN_OPEN)) {
898 /* Well, there's extra open request that we do not need,
899 * let's close it somehow. This will decref request. */
900 rc = it_open_error(DISP_OPEN_OPEN, it);
902 mutex_unlock(&lli->lli_och_mutex);
903 GOTO(out_openerr, rc);
906 ll_release_openhandle(file_dentry(file), it);
910 rc = ll_local_open(file, it, fd, NULL);
913 mutex_unlock(&lli->lli_och_mutex);
914 GOTO(out_openerr, rc);
917 LASSERT(*och_usecount == 0);
918 if (!it->it_disposition) {
919 struct dentry *dentry = file_dentry(file);
920 struct ll_sb_info *sbi = ll_i2sbi(inode);
921 struct ll_dentry_data *ldd;
923 /* We cannot just request lock handle now, new ELC code
924 * means that one of other OPEN locks for this file
925 * could be cancelled, and since blocking ast handler
926 * would attempt to grab och_mutex as well, that would
927 * result in a deadlock
929 mutex_unlock(&lli->lli_och_mutex);
931 * Normally called under two situations:
933 * 2. A race/condition on MDS resulting in no open
934 * handle to be returned from LOOKUP|OPEN request,
935 * for example if the target entry was a symlink.
937 * In NFS path we know there's pathologic behavior
938 * so we always enable open lock caching when coming
939 * from there. It's detected by setting a flag in
942 * After reaching number of opens of this inode
943 * we always ask for an open lock on it to handle
944 * bad userspace actors that open and close files
945 * in a loop for absolutely no good reason
948 ldd = ll_d2d(dentry);
949 if (filename_is_volatile(dentry->d_name.name,
952 /* There really is nothing here, but this
953 * make this more readable I think.
954 * We do not want openlock for volatile
955 * files under any circumstances
957 } else if (ldd && ldd->lld_nfs_dentry) {
958 /* NFS path. This also happens to catch
959 * open by fh files I guess
961 it->it_flags |= MDS_OPEN_LOCK;
962 /* clear the flag for future lookups */
963 ldd->lld_nfs_dentry = 0;
964 } else if (sbi->ll_oc_thrsh_count > 0) {
965 /* Take MDS_OPEN_LOCK with many opens */
966 if (lli->lli_open_fd_count >=
967 sbi->ll_oc_thrsh_count)
968 it->it_flags |= MDS_OPEN_LOCK;
970 /* If this is open after we just closed */
971 else if (ktime_before(ktime_get(),
972 ktime_add_ms(lli->lli_close_fd_time,
973 sbi->ll_oc_thrsh_ms)))
974 it->it_flags |= MDS_OPEN_LOCK;
978 * Always specify MDS_OPEN_BY_FID because we don't want
979 * to get file with different fid.
981 it->it_flags |= MDS_OPEN_BY_FID;
982 rc = ll_intent_file_open(dentry, NULL, 0, it);
984 GOTO(out_openerr, rc);
988 OBD_ALLOC(*och_p, sizeof(struct obd_client_handle));
990 GOTO(out_och_free, rc = -ENOMEM);
994 /* md_intent_lock() didn't get a request ref if there was an
995 * open error, so don't do cleanup on the request here
997 /* XXX (green): Should not we bail out on any error here, not
998 * just open error? */
999 rc = it_open_error(DISP_OPEN_OPEN, it);
1001 GOTO(out_och_free, rc);
1003 LASSERTF(it_disposition(it, DISP_ENQ_OPEN_REF),
1004 "inode %p: disposition %x, status %d\n", inode,
1005 it_disposition(it, ~0), it->it_status);
1007 rc = ll_local_open(file, it, fd, *och_p);
1009 GOTO(out_och_free, rc);
1012 rc = pcc_file_open(inode, file);
1014 GOTO(out_och_free, rc);
1016 mutex_unlock(&lli->lli_och_mutex);
1020 /* Must do this outside lli_och_mutex lock to prevent deadlock where
1021 different kind of OPEN lock for this same inode gets cancelled
1022 by ldlm_cancel_lru */
1023 if (!S_ISREG(inode->i_mode))
1024 GOTO(out_och_free, rc);
1025 cl_lov_delay_create_clear(&file->f_flags);
1026 GOTO(out_och_free, rc);
1030 if (och_p && *och_p) {
1031 OBD_FREE(*och_p, sizeof(struct obd_client_handle));
1032 *och_p = NULL; /* OBD_FREE writes some magic there */
1035 mutex_unlock(&lli->lli_och_mutex);
1038 if (lli->lli_opendir_key == fd)
1039 ll_deauthorize_statahead(inode, fd);
1042 ll_file_data_put(fd);
1044 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN,
1045 ktime_us_delta(ktime_get(), kstart));
1049 if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) {
1050 ptlrpc_req_finished(it->it_request);
1051 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1057 static int ll_md_blocking_lease_ast(struct ldlm_lock *lock,
1058 struct ldlm_lock_desc *desc, void *data, int flag)
1061 struct lustre_handle lockh;
1065 case LDLM_CB_BLOCKING:
1066 ldlm_lock2handle(lock, &lockh);
1067 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
1069 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1073 case LDLM_CB_CANCELING:
1081 * When setting a lease on a file, we take ownership of the lli_mds_*_och
1082 * and save it as fd->fd_och so as to force client to reopen the file even
1083 * if it has an open lock in cache already.
1085 static int ll_lease_och_acquire(struct inode *inode, struct file *file,
1086 struct lustre_handle *old_open_handle)
1088 struct ll_inode_info *lli = ll_i2info(inode);
1089 struct ll_file_data *fd = file->private_data;
1090 struct obd_client_handle **och_p;
1091 __u64 *och_usecount;
1095 /* Get the openhandle of the file */
1096 mutex_lock(&lli->lli_och_mutex);
1097 if (fd->fd_lease_och != NULL)
1098 GOTO(out_unlock, rc = -EBUSY);
1100 if (fd->fd_och == NULL) {
1101 if (file->f_mode & FMODE_WRITE) {
1102 LASSERT(lli->lli_mds_write_och != NULL);
1103 och_p = &lli->lli_mds_write_och;
1104 och_usecount = &lli->lli_open_fd_write_count;
1106 LASSERT(lli->lli_mds_read_och != NULL);
1107 och_p = &lli->lli_mds_read_och;
1108 och_usecount = &lli->lli_open_fd_read_count;
1111 if (*och_usecount > 1)
1112 GOTO(out_unlock, rc = -EBUSY);
1114 fd->fd_och = *och_p;
1119 *old_open_handle = fd->fd_och->och_open_handle;
1123 mutex_unlock(&lli->lli_och_mutex);
1128 * Release ownership on lli_mds_*_och when putting back a file lease.
1130 static int ll_lease_och_release(struct inode *inode, struct file *file)
1132 struct ll_inode_info *lli = ll_i2info(inode);
1133 struct ll_file_data *fd = file->private_data;
1134 struct obd_client_handle **och_p;
1135 struct obd_client_handle *old_och = NULL;
1136 __u64 *och_usecount;
1140 mutex_lock(&lli->lli_och_mutex);
1141 if (file->f_mode & FMODE_WRITE) {
1142 och_p = &lli->lli_mds_write_och;
1143 och_usecount = &lli->lli_open_fd_write_count;
1145 och_p = &lli->lli_mds_read_och;
1146 och_usecount = &lli->lli_open_fd_read_count;
1149 /* The file may have been open by another process (broken lease) so
1150 * *och_p is not NULL. In this case we should simply increase usecount
1153 if (*och_p != NULL) {
1154 old_och = fd->fd_och;
1157 *och_p = fd->fd_och;
1161 mutex_unlock(&lli->lli_och_mutex);
1163 if (old_och != NULL)
1164 rc = ll_close_inode_openhandle(inode, old_och, 0, NULL);
1170 * Acquire a lease and open the file.
1172 static struct obd_client_handle *
1173 ll_lease_open(struct inode *inode, struct file *file, fmode_t fmode,
1176 struct lookup_intent it = { .it_op = IT_OPEN };
1177 struct ll_sb_info *sbi = ll_i2sbi(inode);
1178 struct md_op_data *op_data;
1179 struct ptlrpc_request *req = NULL;
1180 struct lustre_handle old_open_handle = { 0 };
1181 struct obd_client_handle *och = NULL;
1186 if (fmode != FMODE_WRITE && fmode != FMODE_READ)
1187 RETURN(ERR_PTR(-EINVAL));
1190 if (!(fmode & file->f_mode) || (file->f_mode & FMODE_EXEC))
1191 RETURN(ERR_PTR(-EPERM));
1193 rc = ll_lease_och_acquire(inode, file, &old_open_handle);
1195 RETURN(ERR_PTR(rc));
1200 RETURN(ERR_PTR(-ENOMEM));
1202 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
1203 LUSTRE_OPC_ANY, NULL);
1204 if (IS_ERR(op_data))
1205 GOTO(out, rc = PTR_ERR(op_data));
1207 /* To tell the MDT this openhandle is from the same owner */
1208 op_data->op_open_handle = old_open_handle;
1210 it.it_flags = fmode | open_flags;
1211 it.it_flags |= MDS_OPEN_LOCK | MDS_OPEN_BY_FID | MDS_OPEN_LEASE;
1212 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
1213 &ll_md_blocking_lease_ast,
1214 /* LDLM_FL_NO_LRU: To not put the lease lock into LRU list, otherwise
1215 * it can be cancelled which may mislead applications that the lease is
1217 * LDLM_FL_EXCL: Set this flag so that it won't be matched by normal
1218 * open in ll_md_blocking_ast(). Otherwise as ll_md_blocking_lease_ast
1219 * doesn't deal with openhandle, so normal openhandle will be leaked. */
1220 LDLM_FL_NO_LRU | LDLM_FL_EXCL);
1221 ll_finish_md_op_data(op_data);
1222 ptlrpc_req_finished(req);
1224 GOTO(out_release_it, rc);
1226 if (it_disposition(&it, DISP_LOOKUP_NEG))
1227 GOTO(out_release_it, rc = -ENOENT);
1229 rc = it_open_error(DISP_OPEN_OPEN, &it);
1231 GOTO(out_release_it, rc);
1233 LASSERT(it_disposition(&it, DISP_ENQ_OPEN_REF));
1234 rc = ll_och_fill(sbi->ll_md_exp, &it, och);
1236 GOTO(out_release_it, rc);
1238 if (!it_disposition(&it, DISP_OPEN_LEASE)) /* old server? */
1239 GOTO(out_close, rc = -EOPNOTSUPP);
1241 /* already get lease, handle lease lock */
1242 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
1243 if (!it.it_lock_mode ||
1244 !(it.it_lock_bits & MDS_INODELOCK_OPEN)) {
1245 /* open lock must return for lease */
1246 CERROR(DFID "lease granted but no open lock, %d/%llu.\n",
1247 PFID(ll_inode2fid(inode)), it.it_lock_mode,
1249 GOTO(out_close, rc = -EPROTO);
1252 ll_intent_release(&it);
1256 /* Cancel open lock */
1257 if (it.it_lock_mode != 0) {
1258 ldlm_lock_decref_and_cancel(&och->och_lease_handle,
1260 it.it_lock_mode = 0;
1261 och->och_lease_handle.cookie = 0ULL;
1263 rc2 = ll_close_inode_openhandle(inode, och, 0, NULL);
1265 CERROR("%s: error closing file "DFID": %d\n",
1266 sbi->ll_fsname, PFID(&ll_i2info(inode)->lli_fid), rc2);
1267 och = NULL; /* och has been freed in ll_close_inode_openhandle() */
1269 ll_intent_release(&it);
1273 RETURN(ERR_PTR(rc));
1277 * Check whether a layout swap can be done between two inodes.
1279 * \param[in] inode1 First inode to check
1280 * \param[in] inode2 Second inode to check
1282 * \retval 0 on success, layout swap can be performed between both inodes
1283 * \retval negative error code if requirements are not met
1285 static int ll_check_swap_layouts_validity(struct inode *inode1,
1286 struct inode *inode2)
1288 if (!S_ISREG(inode1->i_mode) || !S_ISREG(inode2->i_mode))
1291 if (inode_permission(&init_user_ns, inode1, MAY_WRITE) ||
1292 inode_permission(&init_user_ns, inode2, MAY_WRITE))
1295 if (inode1->i_sb != inode2->i_sb)
1301 static int ll_swap_layouts_close(struct obd_client_handle *och,
1302 struct inode *inode, struct inode *inode2)
1304 const struct lu_fid *fid1 = ll_inode2fid(inode);
1305 const struct lu_fid *fid2;
1309 CDEBUG(D_INODE, "%s: biased close of file "DFID"\n",
1310 ll_i2sbi(inode)->ll_fsname, PFID(fid1));
1312 rc = ll_check_swap_layouts_validity(inode, inode2);
1314 GOTO(out_free_och, rc);
1316 /* We now know that inode2 is a lustre inode */
1317 fid2 = ll_inode2fid(inode2);
1319 rc = lu_fid_cmp(fid1, fid2);
1321 GOTO(out_free_och, rc = -EINVAL);
1323 /* Close the file and {swap,merge} layouts between inode & inode2.
1324 * NB: lease lock handle is released in mdc_close_layout_swap_pack()
1325 * because we still need it to pack l_remote_handle to MDT. */
1326 rc = ll_close_inode_openhandle(inode, och, MDS_CLOSE_LAYOUT_SWAP,
1329 och = NULL; /* freed in ll_close_inode_openhandle() */
1339 * Release lease and close the file.
1340 * It will check if the lease has ever broken.
1342 static int ll_lease_close_intent(struct obd_client_handle *och,
1343 struct inode *inode,
1344 bool *lease_broken, enum mds_op_bias bias,
1347 struct ldlm_lock *lock;
1348 bool cancelled = true;
1352 lock = ldlm_handle2lock(&och->och_lease_handle);
1354 lock_res_and_lock(lock);
1355 cancelled = ldlm_is_cancel(lock);
1356 unlock_res_and_lock(lock);
1357 LDLM_LOCK_PUT(lock);
1360 CDEBUG(D_INODE, "lease for "DFID" broken? %d, bias: %x\n",
1361 PFID(&ll_i2info(inode)->lli_fid), cancelled, bias);
1363 if (lease_broken != NULL)
1364 *lease_broken = cancelled;
1366 if (!cancelled && !bias)
1367 ldlm_cli_cancel(&och->och_lease_handle, 0);
1369 if (cancelled) { /* no need to excute intent */
1374 rc = ll_close_inode_openhandle(inode, och, bias, data);
1378 static int ll_lease_close(struct obd_client_handle *och, struct inode *inode,
1381 return ll_lease_close_intent(och, inode, lease_broken, 0, NULL);
1385 * After lease is taken, send the RPC MDS_REINT_RESYNC to the MDT
1387 static int ll_lease_file_resync(struct obd_client_handle *och,
1388 struct inode *inode, unsigned long arg)
1390 struct ll_sb_info *sbi = ll_i2sbi(inode);
1391 struct md_op_data *op_data;
1392 struct ll_ioc_lease_id ioc;
1393 __u64 data_version_unused;
1397 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1398 LUSTRE_OPC_ANY, NULL);
1399 if (IS_ERR(op_data))
1400 RETURN(PTR_ERR(op_data));
1402 if (copy_from_user(&ioc, (struct ll_ioc_lease_id __user *)arg,
1406 /* before starting file resync, it's necessary to clean up page cache
1407 * in client memory, otherwise once the layout version is increased,
1408 * writing back cached data will be denied the OSTs. */
1409 rc = ll_data_version(inode, &data_version_unused, LL_DV_WR_FLUSH);
1413 op_data->op_lease_handle = och->och_lease_handle;
1414 op_data->op_mirror_id = ioc.lil_mirror_id;
1415 rc = md_file_resync(sbi->ll_md_exp, op_data);
1421 ll_finish_md_op_data(op_data);
1425 int ll_merge_attr(const struct lu_env *env, struct inode *inode)
1427 struct ll_inode_info *lli = ll_i2info(inode);
1428 struct cl_object *obj = lli->lli_clob;
1429 struct cl_attr *attr = vvp_env_thread_attr(env);
1437 ll_inode_size_lock(inode);
1439 /* Merge timestamps the most recently obtained from MDS with
1440 * timestamps obtained from OSTs.
1442 * Do not overwrite atime of inode because it may be refreshed
1443 * by file_accessed() function. If the read was served by cache
1444 * data, there is no RPC to be sent so that atime may not be
1445 * transferred to OSTs at all. MDT only updates atime at close time
1446 * if it's at least 'mdd.*.atime_diff' older.
1447 * All in all, the atime in Lustre does not strictly comply with
1448 * POSIX. Solving this problem needs to send an RPC to MDT for each
1449 * read, this will hurt performance.
1451 if (test_and_clear_bit(LLIF_UPDATE_ATIME, &lli->lli_flags) ||
1452 inode->i_atime.tv_sec < lli->lli_atime)
1453 inode->i_atime.tv_sec = lli->lli_atime;
1455 inode->i_mtime.tv_sec = lli->lli_mtime;
1456 inode->i_ctime.tv_sec = lli->lli_ctime;
1458 mtime = inode->i_mtime.tv_sec;
1459 atime = inode->i_atime.tv_sec;
1460 ctime = inode->i_ctime.tv_sec;
1462 cl_object_attr_lock(obj);
1463 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_MERGE))
1466 rc = cl_object_attr_get(env, obj, attr);
1467 cl_object_attr_unlock(obj);
1470 GOTO(out_size_unlock, rc = (rc == -ENODATA ? 0 : rc));
1472 if (atime < attr->cat_atime)
1473 atime = attr->cat_atime;
1475 if (ctime < attr->cat_ctime)
1476 ctime = attr->cat_ctime;
1478 if (mtime < attr->cat_mtime)
1479 mtime = attr->cat_mtime;
1481 CDEBUG(D_VFSTRACE, DFID" updating i_size %llu\n",
1482 PFID(&lli->lli_fid), attr->cat_size);
1484 if (llcrypt_require_key(inode) == -ENOKEY) {
1485 /* Without the key, round up encrypted file size to next
1486 * LUSTRE_ENCRYPTION_UNIT_SIZE. Clear text size is put in
1487 * lli_lazysize for proper file size setting at close time.
1489 lli->lli_attr_valid |= OBD_MD_FLLAZYSIZE;
1490 lli->lli_lazysize = attr->cat_size;
1491 attr->cat_size = round_up(attr->cat_size,
1492 LUSTRE_ENCRYPTION_UNIT_SIZE);
1494 i_size_write(inode, attr->cat_size);
1495 inode->i_blocks = attr->cat_blocks;
1497 inode->i_mtime.tv_sec = mtime;
1498 inode->i_atime.tv_sec = atime;
1499 inode->i_ctime.tv_sec = ctime;
1502 ll_inode_size_unlock(inode);
1508 * Set designated mirror for I/O.
1510 * So far only read, write, and truncated can support to issue I/O to
1511 * designated mirror.
1513 void ll_io_set_mirror(struct cl_io *io, const struct file *file)
1515 struct ll_file_data *fd = file->private_data;
1517 /* clear layout version for generic(non-resync) I/O in case it carries
1518 * stale layout version due to I/O restart */
1519 io->ci_layout_version = 0;
1521 /* FLR: disable non-delay for designated mirror I/O because obviously
1522 * only one mirror is available */
1523 if (fd->fd_designated_mirror > 0) {
1525 io->ci_designated_mirror = fd->fd_designated_mirror;
1526 io->ci_layout_version = fd->fd_layout_version;
1529 CDEBUG(D_VFSTRACE, "%s: desiginated mirror: %d\n",
1530 file->f_path.dentry->d_name.name, io->ci_designated_mirror);
1533 static bool file_is_noatime(const struct file *file)
1535 const struct vfsmount *mnt = file->f_path.mnt;
1536 const struct inode *inode = file_inode((struct file *)file);
1538 /* Adapted from file_accessed() and touch_atime().*/
1539 if (file->f_flags & O_NOATIME)
1542 if (inode->i_flags & S_NOATIME)
1545 if (IS_NOATIME(inode))
1548 if (mnt->mnt_flags & (MNT_NOATIME | MNT_READONLY))
1551 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1554 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1560 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1561 struct vvp_io_args *args)
1563 struct inode *inode = file_inode(file);
1564 struct ll_file_data *fd = file->private_data;
1566 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
1567 io->ci_lock_no_expand = fd->ll_lock_no_expand;
1569 if (iot == CIT_WRITE) {
1570 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
1571 io->u.ci_wr.wr_sync = !!(file->f_flags & O_SYNC ||
1572 file->f_flags & O_DIRECT ||
1574 #ifdef HAVE_GENERIC_WRITE_SYNC_2ARGS
1575 io->u.ci_wr.wr_sync |= !!(args &&
1576 (args->u.normal.via_iocb->ki_flags &
1581 io->ci_obj = ll_i2info(inode)->lli_clob;
1582 io->ci_lockreq = CILR_MAYBE;
1583 if (ll_file_nolock(file)) {
1584 io->ci_lockreq = CILR_NEVER;
1585 io->ci_no_srvlock = 1;
1586 } else if (file->f_flags & O_APPEND) {
1587 io->ci_lockreq = CILR_MANDATORY;
1589 io->ci_noatime = file_is_noatime(file);
1590 io->ci_async_readahead = false;
1592 /* FLR: only use non-delay I/O for read as there is only one
1593 * avaliable mirror for write. */
1594 io->ci_ndelay = !(iot == CIT_WRITE);
1596 ll_io_set_mirror(io, file);
1599 static void ll_heat_add(struct inode *inode, enum cl_io_type iot,
1602 struct ll_inode_info *lli = ll_i2info(inode);
1603 struct ll_sb_info *sbi = ll_i2sbi(inode);
1604 enum obd_heat_type sample_type;
1605 enum obd_heat_type iobyte_type;
1606 __u64 now = ktime_get_real_seconds();
1608 if (!ll_sbi_has_file_heat(sbi) ||
1609 lli->lli_heat_flags & LU_HEAT_FLAG_OFF)
1612 if (iot == CIT_READ) {
1613 sample_type = OBD_HEAT_READSAMPLE;
1614 iobyte_type = OBD_HEAT_READBYTE;
1615 } else if (iot == CIT_WRITE) {
1616 sample_type = OBD_HEAT_WRITESAMPLE;
1617 iobyte_type = OBD_HEAT_WRITEBYTE;
1622 spin_lock(&lli->lli_heat_lock);
1623 obd_heat_add(&lli->lli_heat_instances[sample_type], now, 1,
1624 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1625 obd_heat_add(&lli->lli_heat_instances[iobyte_type], now, count,
1626 sbi->ll_heat_decay_weight, sbi->ll_heat_period_second);
1627 spin_unlock(&lli->lli_heat_lock);
1631 ll_file_io_generic(const struct lu_env *env, struct vvp_io_args *args,
1632 struct file *file, enum cl_io_type iot,
1633 loff_t *ppos, size_t count)
1635 struct vvp_io *vio = vvp_env_io(env);
1636 struct inode *inode = file_inode(file);
1637 struct ll_inode_info *lli = ll_i2info(inode);
1638 struct ll_sb_info *sbi = ll_i2sbi(inode);
1639 struct ll_file_data *fd = file->private_data;
1640 struct range_lock range;
1641 bool range_locked = false;
1646 unsigned int retried = 0, dio_lock = 0;
1647 bool is_aio = false;
1648 bool is_parallel_dio = false;
1649 struct cl_dio_aio *ci_aio = NULL;
1651 bool partial_io = false;
1652 size_t max_io_pages, max_cached_pages;
1656 CDEBUG(D_VFSTRACE, "%s: %s ppos: %llu, count: %zu\n",
1657 file_dentry(file)->d_name.name,
1658 iot == CIT_READ ? "read" : "write", *ppos, count);
1660 max_io_pages = PTLRPC_MAX_BRW_PAGES * OBD_MAX_RIF_DEFAULT;
1661 max_cached_pages = sbi->ll_cache->ccc_lru_max;
1662 if (max_io_pages > (max_cached_pages >> 2))
1663 max_io_pages = max_cached_pages >> 2;
1665 io = vvp_env_thread_io(env);
1666 if (file->f_flags & O_DIRECT) {
1667 if (file->f_flags & O_APPEND)
1669 if (!is_sync_kiocb(args->u.normal.via_iocb))
1672 /* the kernel does not support AIO on pipes, and parallel DIO
1673 * uses part of the AIO path, so we must not do parallel dio
1676 is_parallel_dio = !iov_iter_is_pipe(args->u.normal.via_iter) &&
1679 if (!ll_sbi_has_parallel_dio(sbi))
1680 is_parallel_dio = false;
1682 ci_aio = cl_aio_alloc(args->u.normal.via_iocb,
1683 ll_i2info(inode)->lli_clob, NULL);
1685 GOTO(out, rc = -ENOMEM);
1690 * IO block size need be aware of cached page limit, otherwise
1691 * if we have small max_cached_mb but large block IO issued, io
1692 * could not be finished and blocked whole client.
1694 if (file->f_flags & O_DIRECT)
1697 per_bytes = min(max_io_pages << PAGE_SHIFT, count);
1698 partial_io = per_bytes < count;
1699 io = vvp_env_thread_io(env);
1700 ll_io_init(io, file, iot, args);
1701 io->ci_aio = ci_aio;
1702 io->ci_dio_lock = dio_lock;
1703 io->ci_ndelay_tried = retried;
1704 io->ci_parallel_dio = is_parallel_dio;
1706 if (cl_io_rw_init(env, io, iot, *ppos, per_bytes) == 0) {
1707 if (file->f_flags & O_APPEND)
1708 range_lock_init(&range, 0, LUSTRE_EOF);
1710 range_lock_init(&range, *ppos, *ppos + per_bytes - 1);
1712 vio->vui_fd = file->private_data;
1713 vio->vui_iter = args->u.normal.via_iter;
1714 vio->vui_iocb = args->u.normal.via_iocb;
1715 /* Direct IO reads must also take range lock,
1716 * or multiple reads will try to work on the same pages
1717 * See LU-6227 for details.
1719 if (((iot == CIT_WRITE) ||
1720 (iot == CIT_READ && (file->f_flags & O_DIRECT))) &&
1721 !(vio->vui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1722 CDEBUG(D_VFSTRACE, "Range lock "RL_FMT"\n",
1724 rc = range_lock(&lli->lli_write_tree, &range);
1728 range_locked = true;
1731 ll_cl_add(inode, env, io, LCC_RW);
1732 rc = cl_io_loop(env, io);
1733 ll_cl_remove(inode, env);
1735 if (range_locked && !is_parallel_dio) {
1736 CDEBUG(D_VFSTRACE, "Range unlock "RL_FMT"\n",
1738 range_unlock(&lli->lli_write_tree, &range);
1739 range_locked = false;
1742 /* cl_io_rw_init() handled IO */
1746 /* N/B: parallel DIO may be disabled during i/o submission;
1747 * if that occurs, async RPCs are resolved before we get here, and this
1748 * wait call completes immediately.
1750 if (is_parallel_dio) {
1751 struct cl_sync_io *anchor = &io->ci_aio->cda_sync;
1753 /* for dio, EIOCBQUEUED is an implementation detail,
1754 * and we don't return it to userspace
1756 if (rc == -EIOCBQUEUED)
1759 rc2 = cl_sync_io_wait_recycle(env, anchor, 0, 0);
1764 range_unlock(&lli->lli_write_tree, &range);
1765 range_locked = false;
1770 * In order to move forward AIO, ci_nob was increased,
1771 * but that doesn't mean io have been finished, it just
1772 * means io have been submited, we will always return
1773 * EIOCBQUEUED to the caller, So we could only return
1774 * number of bytes in non-AIO case.
1776 if (io->ci_nob > 0) {
1779 result += io->ci_nob;
1780 *ppos = io->u.ci_wr.wr.crw_pos; /* for splice */
1785 count -= io->ci_nob;
1787 /* prepare IO restart */
1789 args->u.normal.via_iter = vio->vui_iter;
1793 * Reexpand iov count because it was zero
1796 iov_iter_reexpand(vio->vui_iter, count);
1797 if (per_bytes == io->ci_nob)
1798 io->ci_need_restart = 1;
1802 cl_io_fini(env, io);
1805 "%s: %d io complete with rc: %d, result: %zd, restart: %d\n",
1806 file->f_path.dentry->d_name.name,
1807 iot, rc, result, io->ci_need_restart);
1809 if ((rc == 0 || rc == -ENODATA || rc == -ENOLCK) &&
1810 count > 0 && io->ci_need_restart) {
1812 "%s: restart %s from %lld, count: %zu, ret: %zd, rc: %d\n",
1813 file_dentry(file)->d_name.name,
1814 iot == CIT_READ ? "read" : "write",
1815 *ppos, count, result, rc);
1816 /* preserve the tried count for FLR */
1817 retried = io->ci_ndelay_tried;
1818 dio_lock = io->ci_dio_lock;
1824 * VFS will call aio_complete() if no -EIOCBQUEUED
1825 * is returned for AIO, so we can not call aio_complete()
1828 if (rc != -EIOCBQUEUED)
1829 io->ci_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_aio->cda_sync,
1836 rc == -EIOCBQUEUED ? 0 : rc);
1838 cl_aio_free(env, io->ci_aio);
1843 if (iot == CIT_READ) {
1845 ll_stats_ops_tally(ll_i2sbi(inode),
1846 LPROC_LL_READ_BYTES, result);
1847 } else if (iot == CIT_WRITE) {
1849 ll_stats_ops_tally(ll_i2sbi(inode),
1850 LPROC_LL_WRITE_BYTES, result);
1851 fd->fd_write_failed = false;
1852 } else if (result == 0 && rc == 0) {
1855 fd->fd_write_failed = true;
1857 fd->fd_write_failed = false;
1858 } else if (rc != -ERESTARTSYS) {
1859 fd->fd_write_failed = true;
1863 CDEBUG(D_VFSTRACE, "iot: %d, result: %zd\n", iot, result);
1865 ll_heat_add(inode, iot, result);
1867 RETURN(result > 0 ? result : rc);
1871 * The purpose of fast read is to overcome per I/O overhead and improve IOPS
1872 * especially for small I/O.
1874 * To serve a read request, CLIO has to create and initialize a cl_io and
1875 * then request DLM lock. This has turned out to have siginificant overhead
1876 * and affects the performance of small I/O dramatically.
1878 * It's not necessary to create a cl_io for each I/O. Under the help of read
1879 * ahead, most of the pages being read are already in memory cache and we can
1880 * read those pages directly because if the pages exist, the corresponding DLM
1881 * lock must exist so that page content must be valid.
1883 * In fast read implementation, the llite speculatively finds and reads pages
1884 * in memory cache. There are three scenarios for fast read:
1885 * - If the page exists and is uptodate, kernel VM will provide the data and
1886 * CLIO won't be intervened;
1887 * - If the page was brought into memory by read ahead, it will be exported
1888 * and read ahead parameters will be updated;
1889 * - Otherwise the page is not in memory, we can't do fast read. Therefore,
1890 * it will go back and invoke normal read, i.e., a cl_io will be created
1891 * and DLM lock will be requested.
1893 * POSIX compliance: posix standard states that read is intended to be atomic.
1894 * Lustre read implementation is in line with Linux kernel read implementation
1895 * and neither of them complies with POSIX standard in this matter. Fast read
1896 * doesn't make the situation worse on single node but it may interleave write
1897 * results from multiple nodes due to short read handling in ll_file_aio_read().
1899 * \param env - lu_env
1900 * \param iocb - kiocb from kernel
1901 * \param iter - user space buffers where the data will be copied
1903 * \retval - number of bytes have been read, or error code if error occurred.
1906 ll_do_fast_read(struct kiocb *iocb, struct iov_iter *iter)
1910 if (!ll_sbi_has_fast_read(ll_i2sbi(file_inode(iocb->ki_filp))))
1913 /* NB: we can't do direct IO for fast read because it will need a lock
1914 * to make IO engine happy. */
1915 if (iocb->ki_filp->f_flags & O_DIRECT)
1918 result = generic_file_read_iter(iocb, iter);
1920 /* If the first page is not in cache, generic_file_aio_read() will be
1921 * returned with -ENODATA.
1922 * See corresponding code in ll_readpage(). */
1923 if (result == -ENODATA)
1927 ll_heat_add(file_inode(iocb->ki_filp), CIT_READ, result);
1928 ll_stats_ops_tally(ll_i2sbi(file_inode(iocb->ki_filp)),
1929 LPROC_LL_READ_BYTES, result);
1936 * Read from a file (through the page cache).
1938 static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1941 struct vvp_io_args *args;
1942 struct file *file = iocb->ki_filp;
1946 ktime_t kstart = ktime_get();
1951 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
1952 file_dentry(file)->d_name.name,
1953 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
1954 iov_iter_count(to));
1956 if (!iov_iter_count(to))
1960 * Currently when PCC read failed, we do not fall back to the
1961 * normal read path, just return the error.
1962 * The resaon is that: for RW-PCC, the file data may be modified
1963 * in the PCC and inconsistent with the data on OSTs (or file
1964 * data has been removed from the Lustre file system), at this
1965 * time, fallback to the normal read path may read the wrong
1967 * TODO: for RO-PCC (readonly PCC), fall back to normal read
1968 * path: read data from data copy on OSTs.
1970 result = pcc_file_read_iter(iocb, to, &cached);
1974 ll_ras_enter(file, iocb->ki_pos, iov_iter_count(to));
1976 result = ll_do_fast_read(iocb, to);
1977 if (result < 0 || iov_iter_count(to) == 0)
1980 env = cl_env_get(&refcheck);
1982 RETURN(PTR_ERR(env));
1984 args = ll_env_args(env);
1985 args->u.normal.via_iter = to;
1986 args->u.normal.via_iocb = iocb;
1988 rc2 = ll_file_io_generic(env, args, file, CIT_READ,
1989 &iocb->ki_pos, iov_iter_count(to));
1992 else if (result == 0)
1995 cl_env_put(env, &refcheck);
1998 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
1999 file->private_data, iocb->ki_pos, result,
2001 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_READ,
2002 ktime_us_delta(ktime_get(), kstart));
2006 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2007 file_dentry(file)->d_name.name,
2008 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2009 iov_iter_count(to));
2015 * Similar trick to ll_do_fast_read, this improves write speed for tiny writes.
2016 * If a page is already in the page cache and dirty (and some other things -
2017 * See ll_tiny_write_begin for the instantiation of these rules), then we can
2018 * write to it without doing a full I/O, because Lustre already knows about it
2019 * and will write it out. This saves a lot of processing time.
2021 * All writes here are within one page, so exclusion is handled by the page
2022 * lock on the vm page. We do not do tiny writes for writes which touch
2023 * multiple pages because it's very unlikely multiple sequential pages are
2024 * are already dirty.
2026 * We limit these to < PAGE_SIZE because PAGE_SIZE writes are relatively common
2027 * and are unlikely to be to already dirty pages.
2029 * Attribute updates are important here, we do them in ll_tiny_write_end.
2031 static ssize_t ll_do_tiny_write(struct kiocb *iocb, struct iov_iter *iter)
2033 ssize_t count = iov_iter_count(iter);
2034 struct file *file = iocb->ki_filp;
2035 struct inode *inode = file_inode(file);
2036 bool lock_inode = !IS_NOSEC(inode);
2041 /* Restrict writes to single page and < PAGE_SIZE. See comment at top
2042 * of function for why.
2044 if (count >= PAGE_SIZE ||
2045 (iocb->ki_pos & (PAGE_SIZE-1)) + count > PAGE_SIZE)
2048 if (unlikely(lock_inode))
2050 result = __generic_file_write_iter(iocb, iter);
2052 if (unlikely(lock_inode))
2053 inode_unlock(inode);
2055 /* If the page is not already dirty, ll_tiny_write_begin returns
2056 * -ENODATA. We continue on to normal write.
2058 if (result == -ENODATA)
2062 ll_heat_add(inode, CIT_WRITE, result);
2063 set_bit(LLIF_DATA_MODIFIED, &ll_i2info(inode)->lli_flags);
2066 CDEBUG(D_VFSTRACE, "result: %zu, original count %zu\n", result, count);
2072 * Write to a file (through the page cache).
2074 static ssize_t ll_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2076 struct vvp_io_args *args;
2078 ssize_t rc_tiny = 0, rc_normal;
2079 struct file *file = iocb->ki_filp;
2082 ktime_t kstart = ktime_get();
2087 CDEBUG(D_VFSTRACE|D_IOTRACE, "file %s:"DFID", ppos: %lld, count: %zu\n",
2088 file_dentry(file)->d_name.name,
2089 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2090 iov_iter_count(from));
2092 if (!iov_iter_count(from))
2093 GOTO(out, rc_normal = 0);
2096 * When PCC write failed, we usually do not fall back to the normal
2097 * write path, just return the error. But there is a special case when
2098 * returned error code is -ENOSPC due to running out of space on PCC HSM
2099 * bakcend. At this time, it will fall back to normal I/O path and
2100 * retry the I/O. As the file is in HSM released state, it will restore
2101 * the file data to OSTs first and redo the write again. And the
2102 * restore process will revoke the layout lock and detach the file
2103 * from PCC cache automatically.
2105 result = pcc_file_write_iter(iocb, from, &cached);
2106 if (cached && result != -ENOSPC && result != -EDQUOT)
2107 GOTO(out, rc_normal = result);
2109 /* NB: we can't do direct IO for tiny writes because they use the page
2110 * cache, we can't do sync writes because tiny writes can't flush
2111 * pages, and we can't do append writes because we can't guarantee the
2112 * required DLM locks are held to protect file size.
2114 if (ll_sbi_has_tiny_write(ll_i2sbi(file_inode(file))) &&
2115 !(file->f_flags & (O_DIRECT | O_SYNC | O_APPEND)))
2116 rc_tiny = ll_do_tiny_write(iocb, from);
2118 /* In case of error, go on and try normal write - Only stop if tiny
2119 * write completed I/O.
2121 if (iov_iter_count(from) == 0)
2122 GOTO(out, rc_normal = rc_tiny);
2124 env = cl_env_get(&refcheck);
2126 RETURN(PTR_ERR(env));
2128 args = ll_env_args(env);
2129 args->u.normal.via_iter = from;
2130 args->u.normal.via_iocb = iocb;
2132 rc_normal = ll_file_io_generic(env, args, file, CIT_WRITE,
2133 &iocb->ki_pos, iov_iter_count(from));
2135 /* On success, combine bytes written. */
2136 if (rc_tiny >= 0 && rc_normal > 0)
2137 rc_normal += rc_tiny;
2138 /* On error, only return error from normal write if tiny write did not
2139 * write any bytes. Otherwise return bytes written by tiny write.
2141 else if (rc_tiny > 0)
2142 rc_normal = rc_tiny;
2144 cl_env_put(env, &refcheck);
2146 if (rc_normal > 0) {
2147 ll_rw_stats_tally(ll_i2sbi(file_inode(file)), current->pid,
2148 file->private_data, iocb->ki_pos,
2150 ll_stats_ops_tally(ll_i2sbi(file_inode(file)), LPROC_LL_WRITE,
2151 ktime_us_delta(ktime_get(), kstart));
2155 "COMPLETED: file %s:"DFID", ppos: %lld, count: %zu\n",
2156 file_dentry(file)->d_name.name,
2157 PFID(ll_inode2fid(file_inode(file))), iocb->ki_pos,
2158 iov_iter_count(from));
2163 #ifndef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
2165 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
2167 static int ll_file_get_iov_count(const struct iovec *iov,
2168 unsigned long *nr_segs, size_t *count,
2174 for (seg = 0; seg < *nr_segs; seg++) {
2175 const struct iovec *iv = &iov[seg];
2178 * If any segment has a negative length, or the cumulative
2179 * length ever wraps negative then return -EINVAL.
2182 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
2184 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
2189 cnt -= iv->iov_len; /* This segment is no good */
2196 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2197 unsigned long nr_segs, loff_t pos)
2204 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_READ);
2211 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2212 iov_iter_init(&to, READ, iov, nr_segs, iov_count);
2213 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2214 iov_iter_init(&to, iov, nr_segs, iov_count, 0);
2215 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2217 result = ll_file_read_iter(iocb, &to);
2222 static ssize_t ll_file_read(struct file *file, char __user *buf, size_t count,
2225 struct iovec iov = { .iov_base = buf, .iov_len = count };
2234 init_sync_kiocb(&kiocb, file);
2235 kiocb.ki_pos = *ppos;
2236 #ifdef HAVE_KIOCB_KI_LEFT
2237 kiocb.ki_left = count;
2238 #elif defined(HAVE_KI_NBYTES)
2239 kiocb.i_nbytes = count;
2242 result = ll_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
2243 *ppos = kiocb.ki_pos;
2249 * Write to a file (through the page cache).
2252 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2253 unsigned long nr_segs, loff_t pos)
2255 struct iov_iter from;
2260 result = ll_file_get_iov_count(iov, &nr_segs, &iov_count, VERIFY_WRITE);
2267 # ifdef HAVE_IOV_ITER_INIT_DIRECTION
2268 iov_iter_init(&from, WRITE, iov, nr_segs, iov_count);
2269 # else /* !HAVE_IOV_ITER_INIT_DIRECTION */
2270 iov_iter_init(&from, iov, nr_segs, iov_count, 0);
2271 # endif /* HAVE_IOV_ITER_INIT_DIRECTION */
2273 result = ll_file_write_iter(iocb, &from);
2278 static ssize_t ll_file_write(struct file *file, const char __user *buf,
2279 size_t count, loff_t *ppos)
2281 struct iovec iov = { .iov_base = (void __user *)buf,
2291 init_sync_kiocb(&kiocb, file);
2292 kiocb.ki_pos = *ppos;
2293 #ifdef HAVE_KIOCB_KI_LEFT
2294 kiocb.ki_left = count;
2295 #elif defined(HAVE_KI_NBYTES)
2296 kiocb.ki_nbytes = count;
2299 result = ll_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
2300 *ppos = kiocb.ki_pos;
2304 #endif /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
2306 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
2307 __u64 flags, struct lov_user_md *lum, int lum_size)
2309 struct lookup_intent oit = {
2311 .it_flags = flags | MDS_OPEN_BY_FID,
2316 if ((__swab32(lum->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
2317 le32_to_cpu(LOV_MAGIC_MAGIC)) {
2318 /* this code will only exist for big-endian systems */
2319 lustre_swab_lov_user_md(lum, 0);
2322 ll_inode_size_lock(inode);
2323 rc = ll_intent_file_open(dentry, lum, lum_size, &oit);
2325 GOTO(out_unlock, rc);
2327 ll_release_openhandle(dentry, &oit);
2330 ll_inode_size_unlock(inode);
2331 ll_intent_release(&oit);
2336 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
2337 struct lov_mds_md **lmmp, int *lmm_size,
2338 struct ptlrpc_request **request)
2340 struct ll_sb_info *sbi = ll_i2sbi(inode);
2341 struct mdt_body *body;
2342 struct lov_mds_md *lmm = NULL;
2343 struct ptlrpc_request *req = NULL;
2344 struct md_op_data *op_data;
2349 rc = ll_get_default_mdsize(sbi, &lmmsize);
2353 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
2354 strlen(filename), lmmsize,
2355 LUSTRE_OPC_ANY, NULL);
2356 if (IS_ERR(op_data))
2357 RETURN(PTR_ERR(op_data));
2359 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
2360 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
2361 ll_finish_md_op_data(op_data);
2363 CDEBUG(D_INFO, "md_getattr_name failed on %s: rc %d\n",
2368 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2369 LASSERT(body != NULL); /* checked by mdc_getattr_name */
2371 lmmsize = body->mbo_eadatasize;
2373 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
2375 GOTO(out, rc = -ENODATA);
2377 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
2378 LASSERT(lmm != NULL);
2380 if (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1) &&
2381 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3) &&
2382 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_COMP_V1) &&
2383 lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_FOREIGN))
2384 GOTO(out, rc = -EPROTO);
2387 * This is coming from the MDS, so is probably in
2388 * little endian. We convert it to host endian before
2389 * passing it to userspace.
2391 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
2392 int stripe_count = 0;
2394 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
2395 lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
2396 stripe_count = le16_to_cpu(lmm->lmm_stripe_count);
2397 if (le32_to_cpu(lmm->lmm_pattern) &
2398 LOV_PATTERN_F_RELEASED)
2400 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
2402 /* if function called for directory - we should
2403 * avoid swab not existent lsm objects
2405 if (lmm->lmm_magic == LOV_MAGIC_V1 &&
2406 S_ISREG(body->mbo_mode))
2407 lustre_swab_lov_user_md_objects(
2408 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
2410 else if (lmm->lmm_magic == LOV_MAGIC_V3 &&
2411 S_ISREG(body->mbo_mode))
2412 lustre_swab_lov_user_md_objects(
2413 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
2415 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
2416 lustre_swab_lov_comp_md_v1(
2417 (struct lov_comp_md_v1 *)lmm);
2421 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
2422 struct lov_comp_md_v1 *comp_v1 = NULL;
2423 struct lov_comp_md_entry_v1 *ent;
2424 struct lov_user_md_v1 *v1;
2428 comp_v1 = (struct lov_comp_md_v1 *)lmm;
2429 /* Dump the striping information */
2430 for (; i < comp_v1->lcm_entry_count; i++) {
2431 ent = &comp_v1->lcm_entries[i];
2432 off = ent->lcme_offset;
2433 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2435 "comp[%d]: stripe_count=%u, stripe_size=%u\n",
2436 i, v1->lmm_stripe_count, v1->lmm_stripe_size);
2440 * Return valid stripe_count and stripe_size instead of 0 for
2441 * DoM files to avoid divide-by-zero for older userspace that
2442 * calls this ioctl, e.g. lustre ADIO driver.
2444 if (lmm->lmm_stripe_count == 0)
2445 lmm->lmm_stripe_count = 1;
2446 if (lmm->lmm_stripe_size == 0) {
2447 /* Since the first component of the file data is placed
2448 * on the MDT for faster access, the stripe_size of the
2449 * second one is always that applications which are
2452 if (lmm->lmm_pattern == LOV_PATTERN_MDT)
2453 i = comp_v1->lcm_entry_count > 1 ? 1 : 0;
2455 i = comp_v1->lcm_entry_count > 1 ?
2456 comp_v1->lcm_entry_count - 1 : 0;
2457 ent = &comp_v1->lcm_entries[i];
2458 off = ent->lcme_offset;
2459 v1 = (struct lov_user_md_v1 *)((char *)lmm + off);
2460 lmm->lmm_stripe_size = v1->lmm_stripe_size;
2465 *lmm_size = lmmsize;
2470 static int ll_lov_setea(struct inode *inode, struct file *file,
2473 __u64 flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
2474 struct lov_user_md *lump;
2475 int lum_size = sizeof(struct lov_user_md) +
2476 sizeof(struct lov_user_ost_data);
2480 if (!capable(CAP_SYS_ADMIN))
2483 OBD_ALLOC_LARGE(lump, lum_size);
2487 if (copy_from_user(lump, arg, lum_size))
2488 GOTO(out_lump, rc = -EFAULT);
2490 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, lump,
2492 cl_lov_delay_create_clear(&file->f_flags);
2495 OBD_FREE_LARGE(lump, lum_size);
2499 static int ll_file_getstripe(struct inode *inode, void __user *lum, size_t size)
2506 env = cl_env_get(&refcheck);
2508 RETURN(PTR_ERR(env));
2510 rc = cl_object_getstripe(env, ll_i2info(inode)->lli_clob, lum, size);
2511 cl_env_put(env, &refcheck);
2515 static int ll_lov_setstripe(struct inode *inode, struct file *file,
2518 struct lov_user_md __user *lum = (struct lov_user_md __user *)arg;
2519 struct lov_user_md *klum;
2521 __u64 flags = FMODE_WRITE;
2524 rc = ll_copy_user_md(lum, &klum);
2529 rc = ll_lov_setstripe_ea_info(inode, file_dentry(file), flags, klum,
2534 rc = put_user(0, &lum->lmm_stripe_count);
2538 rc = ll_layout_refresh(inode, &gen);
2542 rc = ll_file_getstripe(inode, arg, lum_size);
2543 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2544 ll_i2info(inode)->lli_clob) {
2545 struct iattr attr = { 0 };
2547 rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, &attr,
2548 OP_XVALID_FLAGS, LUSTRE_ENCRYPT_FL);
2551 cl_lov_delay_create_clear(&file->f_flags);
2554 OBD_FREE_LARGE(klum, lum_size);
2560 ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
2562 struct ll_inode_info *lli = ll_i2info(inode);
2563 struct cl_object *obj = lli->lli_clob;
2564 struct ll_file_data *fd = file->private_data;
2565 struct ll_grouplock grouplock;
2570 CWARN("group id for group lock must not be 0\n");
2574 if (ll_file_nolock(file))
2575 RETURN(-EOPNOTSUPP);
2577 if (file->f_flags & O_NONBLOCK) {
2578 if (!mutex_trylock(&lli->lli_group_mutex))
2581 mutex_lock(&lli->lli_group_mutex);
2583 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
2584 CWARN("group lock already existed with gid %lu\n",
2585 fd->fd_grouplock.lg_gid);
2586 GOTO(out, rc = -EINVAL);
2588 if (arg != lli->lli_group_gid && lli->lli_group_users != 0) {
2589 if (file->f_flags & O_NONBLOCK)
2590 GOTO(out, rc = -EAGAIN);
2591 mutex_unlock(&lli->lli_group_mutex);
2592 wait_var_event(&lli->lli_group_users, !lli->lli_group_users);
2593 GOTO(retry, rc = 0);
2595 LASSERT(fd->fd_grouplock.lg_lock == NULL);
2598 * XXX: group lock needs to protect all OST objects while PFL
2599 * can add new OST objects during the IO, so we'd instantiate
2600 * all OST objects before getting its group lock.
2605 struct cl_layout cl = {
2606 .cl_is_composite = false,
2608 struct lu_extent ext = {
2610 .e_end = OBD_OBJECT_EOF,
2613 env = cl_env_get(&refcheck);
2615 GOTO(out, rc = PTR_ERR(env));
2617 rc = cl_object_layout_get(env, obj, &cl);
2618 if (rc >= 0 && cl.cl_is_composite)
2619 rc = ll_layout_write_intent(inode, LAYOUT_INTENT_WRITE,
2622 cl_env_put(env, &refcheck);
2627 rc = cl_get_grouplock(ll_i2info(inode)->lli_clob,
2628 arg, (file->f_flags & O_NONBLOCK), &grouplock);
2633 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
2634 fd->fd_grouplock = grouplock;
2635 if (lli->lli_group_users == 0)
2636 lli->lli_group_gid = grouplock.lg_gid;
2637 lli->lli_group_users++;
2639 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
2641 mutex_unlock(&lli->lli_group_mutex);
2646 static int ll_put_grouplock(struct inode *inode, struct file *file,
2649 struct ll_inode_info *lli = ll_i2info(inode);
2650 struct ll_file_data *fd = file->private_data;
2651 struct ll_grouplock grouplock;
2655 mutex_lock(&lli->lli_group_mutex);
2656 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
2657 CWARN("no group lock held\n");
2658 GOTO(out, rc = -EINVAL);
2661 LASSERT(fd->fd_grouplock.lg_lock != NULL);
2663 if (fd->fd_grouplock.lg_gid != arg) {
2664 CWARN("group lock %lu doesn't match current id %lu\n",
2665 arg, fd->fd_grouplock.lg_gid);
2666 GOTO(out, rc = -EINVAL);
2669 grouplock = fd->fd_grouplock;
2670 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
2671 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
2673 cl_put_grouplock(&grouplock);
2675 lli->lli_group_users--;
2676 if (lli->lli_group_users == 0) {
2677 lli->lli_group_gid = 0;
2678 wake_up_var(&lli->lli_group_users);
2680 CDEBUG(D_INFO, "group lock %lu released\n", arg);
2683 mutex_unlock(&lli->lli_group_mutex);
2689 * Close inode open handle
2691 * \param dentry [in] dentry which contains the inode
2692 * \param it [in,out] intent which contains open info and result
2695 * \retval <0 failure
2697 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2699 struct inode *inode = dentry->d_inode;
2700 struct obd_client_handle *och;
2706 /* Root ? Do nothing. */
2707 if (is_root_inode(inode))
2710 /* No open handle to close? Move away */
2711 if (!it_disposition(it, DISP_OPEN_OPEN))
2714 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2716 OBD_ALLOC(och, sizeof(*och));
2718 GOTO(out, rc = -ENOMEM);
2720 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
2724 rc = ll_close_inode_openhandle(inode, och, 0, NULL);
2726 /* this one is in place of ll_file_open */
2727 if (it_disposition(it, DISP_ENQ_OPEN_REF)) {
2728 ptlrpc_req_finished(it->it_request);
2729 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2735 * Get size for inode for which FIEMAP mapping is requested.
2736 * Make the FIEMAP get_info call and returns the result.
2737 * \param fiemap kernel buffer to hold extens
2738 * \param num_bytes kernel buffer size
2740 static int ll_do_fiemap(struct inode *inode, struct fiemap *fiemap,
2746 struct ll_fiemap_info_key fmkey = { .lfik_name = KEY_FIEMAP, };
2749 /* Checks for fiemap flags */
2750 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
2751 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
2755 /* Check for FIEMAP_FLAG_SYNC */
2756 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
2757 rc = filemap_fdatawrite(inode->i_mapping);
2762 env = cl_env_get(&refcheck);
2764 RETURN(PTR_ERR(env));
2766 if (i_size_read(inode) == 0) {
2767 rc = ll_glimpse_size(inode);
2772 fmkey.lfik_oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLPROJID;
2773 obdo_from_inode(&fmkey.lfik_oa, inode, OBD_MD_FLSIZE);
2774 obdo_set_parent_fid(&fmkey.lfik_oa, &ll_i2info(inode)->lli_fid);
2776 /* If filesize is 0, then there would be no objects for mapping */
2777 if (fmkey.lfik_oa.o_size == 0) {
2778 fiemap->fm_mapped_extents = 0;
2782 fmkey.lfik_fiemap = *fiemap;
2784 rc = cl_object_fiemap(env, ll_i2info(inode)->lli_clob,
2785 &fmkey, fiemap, &num_bytes);
2787 cl_env_put(env, &refcheck);
2791 int ll_fid2path(struct inode *inode, void __user *arg)
2793 struct obd_export *exp = ll_i2mdexp(inode);
2794 const struct getinfo_fid2path __user *gfin = arg;
2796 struct getinfo_fid2path *gfout;
2802 if (!capable(CAP_DAC_READ_SEARCH) &&
2803 !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
2806 /* Only need to get the buflen */
2807 if (get_user(pathlen, &gfin->gf_pathlen))
2810 if (pathlen > PATH_MAX)
2813 outsize = sizeof(*gfout) + pathlen;
2814 OBD_ALLOC(gfout, outsize);
2818 if (copy_from_user(gfout, arg, sizeof(*gfout)))
2819 GOTO(gf_free, rc = -EFAULT);
2820 /* append root FID after gfout to let MDT know the root FID so that it
2821 * can lookup the correct path, this is mainly for fileset.
2822 * old server without fileset mount support will ignore this. */
2823 *gfout->gf_u.gf_root_fid = *ll_inode2fid(inode);
2825 /* Call mdc_iocontrol */
2826 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
2830 if (copy_to_user(arg, gfout, outsize))
2834 OBD_FREE(gfout, outsize);
2839 ll_ioc_data_version(struct inode *inode, struct ioc_data_version *ioc)
2841 struct cl_object *obj = ll_i2info(inode)->lli_clob;
2849 ioc->idv_version = 0;
2850 ioc->idv_layout_version = UINT_MAX;
2852 /* If no file object initialized, we consider its version is 0. */
2856 env = cl_env_get(&refcheck);
2858 RETURN(PTR_ERR(env));
2860 io = vvp_env_thread_io(env);
2862 io->u.ci_data_version.dv_data_version = 0;
2863 io->u.ci_data_version.dv_layout_version = UINT_MAX;
2864 io->u.ci_data_version.dv_flags = ioc->idv_flags;
2867 if (cl_io_init(env, io, CIT_DATA_VERSION, io->ci_obj) == 0)
2868 result = cl_io_loop(env, io);
2870 result = io->ci_result;
2872 ioc->idv_version = io->u.ci_data_version.dv_data_version;
2873 ioc->idv_layout_version = io->u.ci_data_version.dv_layout_version;
2875 cl_io_fini(env, io);
2877 if (unlikely(io->ci_need_restart))
2880 cl_env_put(env, &refcheck);
2886 * Read the data_version for inode.
2888 * This value is computed using stripe object version on OST.
2889 * Version is computed using server side locking.
2891 * @param flags if do sync on the OST side;
2893 * LL_DV_RD_FLUSH: flush dirty pages, LCK_PR on OSTs
2894 * LL_DV_WR_FLUSH: drop all caching pages, LCK_PW on OSTs
2896 int ll_data_version(struct inode *inode, __u64 *data_version, int flags)
2898 struct ioc_data_version ioc = { .idv_flags = flags };
2901 rc = ll_ioc_data_version(inode, &ioc);
2903 *data_version = ioc.idv_version;
2909 * Trigger a HSM release request for the provided inode.
2911 int ll_hsm_release(struct inode *inode)
2914 struct obd_client_handle *och = NULL;
2915 __u64 data_version = 0;
2921 CDEBUG(D_INODE, "%s: Releasing file "DFID".\n",
2922 ll_i2sbi(inode)->ll_fsname,
2923 PFID(&ll_i2info(inode)->lli_fid));
2925 och = ll_lease_open(inode, NULL, FMODE_WRITE, MDS_OPEN_RELEASE);
2927 GOTO(out, rc = PTR_ERR(och));
2929 /* Grab latest data_version and [am]time values */
2930 rc = ll_data_version(inode, &data_version,
2931 LL_DV_WR_FLUSH | LL_DV_SZ_UPDATE);
2935 env = cl_env_get(&refcheck);
2937 GOTO(out, rc = PTR_ERR(env));
2939 rc = ll_merge_attr(env, inode);
2940 cl_env_put(env, &refcheck);
2942 /* If error happen, we have the wrong size for a file.
2948 /* Release the file.
2949 * NB: lease lock handle is released in mdc_hsm_release_pack() because
2950 * we still need it to pack l_remote_handle to MDT. */
2951 rc = ll_close_inode_openhandle(inode, och, MDS_HSM_RELEASE,
2957 if (och != NULL && !IS_ERR(och)) /* close the file */
2958 ll_lease_close(och, inode, NULL);
2963 struct ll_swap_stack {
2966 struct inode *inode1;
2967 struct inode *inode2;
2972 static int ll_swap_layouts(struct file *file1, struct file *file2,
2973 struct lustre_swap_layouts *lsl)
2975 struct mdc_swap_layouts msl;
2976 struct md_op_data *op_data;
2979 struct ll_swap_stack *llss = NULL;
2982 OBD_ALLOC_PTR(llss);
2986 llss->inode1 = file_inode(file1);
2987 llss->inode2 = file_inode(file2);
2989 rc = ll_check_swap_layouts_validity(llss->inode1, llss->inode2);
2993 /* we use 2 bool because it is easier to swap than 2 bits */
2994 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV1)
2995 llss->check_dv1 = true;
2997 if (lsl->sl_flags & SWAP_LAYOUTS_CHECK_DV2)
2998 llss->check_dv2 = true;
3000 /* we cannot use lsl->sl_dvX directly because we may swap them */
3001 llss->dv1 = lsl->sl_dv1;
3002 llss->dv2 = lsl->sl_dv2;
3004 rc = lu_fid_cmp(ll_inode2fid(llss->inode1), ll_inode2fid(llss->inode2));
3005 if (rc == 0) /* same file, done! */
3008 if (rc < 0) { /* sequentialize it */
3009 swap(llss->inode1, llss->inode2);
3011 swap(llss->dv1, llss->dv2);
3012 swap(llss->check_dv1, llss->check_dv2);
3016 if (gid != 0) { /* application asks to flush dirty cache */
3017 rc = ll_get_grouplock(llss->inode1, file1, gid);
3021 rc = ll_get_grouplock(llss->inode2, file2, gid);
3023 ll_put_grouplock(llss->inode1, file1, gid);
3028 /* ultimate check, before swaping the layouts we check if
3029 * dataversion has changed (if requested) */
3030 if (llss->check_dv1) {
3031 rc = ll_data_version(llss->inode1, &dv, 0);
3034 if (dv != llss->dv1)
3035 GOTO(putgl, rc = -EAGAIN);
3038 if (llss->check_dv2) {
3039 rc = ll_data_version(llss->inode2, &dv, 0);
3042 if (dv != llss->dv2)
3043 GOTO(putgl, rc = -EAGAIN);
3046 /* struct md_op_data is used to send the swap args to the mdt
3047 * only flags is missing, so we use struct mdc_swap_layouts
3048 * through the md_op_data->op_data */
3049 /* flags from user space have to be converted before they are send to
3050 * server, no flag is sent today, they are only used on the client */
3053 op_data = ll_prep_md_op_data(NULL, llss->inode1, llss->inode2, NULL, 0,
3054 0, LUSTRE_OPC_ANY, &msl);
3055 if (IS_ERR(op_data))
3056 GOTO(free, rc = PTR_ERR(op_data));
3058 rc = obd_iocontrol(LL_IOC_LOV_SWAP_LAYOUTS, ll_i2mdexp(llss->inode1),
3059 sizeof(*op_data), op_data, NULL);
3060 ll_finish_md_op_data(op_data);
3067 ll_put_grouplock(llss->inode2, file2, gid);
3068 ll_put_grouplock(llss->inode1, file1, gid);
3078 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss)
3080 struct obd_export *exp = ll_i2mdexp(inode);
3081 struct md_op_data *op_data;
3085 /* Detect out-of range masks */
3086 if ((hss->hss_setmask | hss->hss_clearmask) & ~HSM_FLAGS_MASK)
3089 /* Non-root users are forbidden to set or clear flags which are
3090 * NOT defined in HSM_USER_MASK. */
3091 if (((hss->hss_setmask | hss->hss_clearmask) & ~HSM_USER_MASK) &&
3092 !capable(CAP_SYS_ADMIN))
3095 if (!exp_connect_archive_id_array(exp)) {
3096 /* Detect out-of range archive id */
3097 if ((hss->hss_valid & HSS_ARCHIVE_ID) &&
3098 (hss->hss_archive_id > LL_HSM_ORIGIN_MAX_ARCHIVE))
3102 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3103 LUSTRE_OPC_ANY, hss);
3104 if (IS_ERR(op_data))
3105 RETURN(PTR_ERR(op_data));
3107 rc = obd_iocontrol(LL_IOC_HSM_STATE_SET, exp, sizeof(*op_data),
3110 ll_finish_md_op_data(op_data);
3115 static int ll_hsm_import(struct inode *inode, struct file *file,
3116 struct hsm_user_import *hui)
3118 struct hsm_state_set *hss = NULL;
3119 struct iattr *attr = NULL;
3123 if (!S_ISREG(inode->i_mode))
3129 GOTO(out, rc = -ENOMEM);
3131 hss->hss_valid = HSS_SETMASK | HSS_ARCHIVE_ID;
3132 hss->hss_archive_id = hui->hui_archive_id;
3133 hss->hss_setmask = HS_ARCHIVED | HS_EXISTS | HS_RELEASED;
3134 rc = ll_hsm_state_set(inode, hss);
3138 OBD_ALLOC_PTR(attr);
3140 GOTO(out, rc = -ENOMEM);
3142 attr->ia_mode = hui->hui_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
3143 attr->ia_mode |= S_IFREG;
3144 attr->ia_uid = make_kuid(&init_user_ns, hui->hui_uid);
3145 attr->ia_gid = make_kgid(&init_user_ns, hui->hui_gid);
3146 attr->ia_size = hui->hui_size;
3147 attr->ia_mtime.tv_sec = hui->hui_mtime;
3148 attr->ia_mtime.tv_nsec = hui->hui_mtime_ns;
3149 attr->ia_atime.tv_sec = hui->hui_atime;
3150 attr->ia_atime.tv_nsec = hui->hui_atime_ns;
3152 attr->ia_valid = ATTR_SIZE | ATTR_MODE | ATTR_FORCE |
3153 ATTR_UID | ATTR_GID |
3154 ATTR_MTIME | ATTR_MTIME_SET |
3155 ATTR_ATIME | ATTR_ATIME_SET;
3159 rc = ll_setattr_raw(file_dentry(file), attr, 0, true);
3163 inode_unlock(inode);
3175 static inline long ll_lease_type_from_fmode(fmode_t fmode)
3177 return ((fmode & FMODE_READ) ? LL_LEASE_RDLCK : 0) |
3178 ((fmode & FMODE_WRITE) ? LL_LEASE_WRLCK : 0);
3181 static int ll_file_futimes_3(struct file *file, const struct ll_futimes_3 *lfu)
3183 struct inode *inode = file_inode(file);
3185 .ia_valid = ATTR_ATIME | ATTR_ATIME_SET |
3186 ATTR_MTIME | ATTR_MTIME_SET |
3189 .tv_sec = lfu->lfu_atime_sec,
3190 .tv_nsec = lfu->lfu_atime_nsec,
3193 .tv_sec = lfu->lfu_mtime_sec,
3194 .tv_nsec = lfu->lfu_mtime_nsec,
3197 .tv_sec = lfu->lfu_ctime_sec,
3198 .tv_nsec = lfu->lfu_ctime_nsec,
3204 if (!capable(CAP_SYS_ADMIN))
3207 if (!S_ISREG(inode->i_mode))
3211 rc = ll_setattr_raw(file_dentry(file), &ia, OP_XVALID_CTIME_SET,
3213 inode_unlock(inode);
3218 static enum cl_lock_mode cl_mode_user_to_kernel(enum lock_mode_user mode)
3221 case MODE_READ_USER:
3223 case MODE_WRITE_USER:
3230 static const char *const user_lockname[] = LOCK_MODE_NAMES;
3232 /* Used to allow the upper layers of the client to request an LDLM lock
3233 * without doing an actual read or write.
3235 * Used for ladvise lockahead to manually request specific locks.
3237 * \param[in] file file this ladvise lock request is on
3238 * \param[in] ladvise ladvise struct describing this lock request
3240 * \retval 0 success, no detailed result available (sync requests
3241 * and requests sent to the server [not handled locally]
3242 * cannot return detailed results)
3243 * \retval LLA_RESULT_{SAME,DIFFERENT} - detailed result of the lock request,
3244 * see definitions for details.
3245 * \retval negative negative errno on error
3247 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise)
3249 struct lu_env *env = NULL;
3250 struct cl_io *io = NULL;
3251 struct cl_lock *lock = NULL;
3252 struct cl_lock_descr *descr = NULL;
3253 struct dentry *dentry = file->f_path.dentry;
3254 struct inode *inode = dentry->d_inode;
3255 enum cl_lock_mode cl_mode;
3256 off_t start = ladvise->lla_start;
3257 off_t end = ladvise->lla_end;
3264 "Lock request: file=%pd, inode=%p, mode=%s start=%llu, end=%llu\n",
3265 dentry, dentry->d_inode,
3266 user_lockname[ladvise->lla_lockahead_mode], (__u64) start,
3269 cl_mode = cl_mode_user_to_kernel(ladvise->lla_lockahead_mode);
3271 GOTO(out, result = cl_mode);
3273 /* Get IO environment */
3274 result = cl_io_get(inode, &env, &io, &refcheck);
3278 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
3281 * nothing to do for this io. This currently happens when
3282 * stripe sub-object's are not yet created.
3284 result = io->ci_result;
3285 } else if (result == 0) {
3286 lock = vvp_env_lock(env);
3287 descr = &lock->cll_descr;
3289 descr->cld_obj = io->ci_obj;
3290 /* Convert byte offsets to pages */
3291 descr->cld_start = cl_index(io->ci_obj, start);
3292 descr->cld_end = cl_index(io->ci_obj, end);
3293 descr->cld_mode = cl_mode;
3294 /* CEF_MUST is used because we do not want to convert a
3295 * lockahead request to a lockless lock */
3296 descr->cld_enq_flags = CEF_MUST | CEF_LOCK_NO_EXPAND;
3298 if (ladvise->lla_peradvice_flags & LF_ASYNC)
3299 descr->cld_enq_flags |= CEF_SPECULATIVE;
3301 result = cl_lock_request(env, io, lock);
3303 /* On success, we need to release the lock */
3305 cl_lock_release(env, lock);
3307 cl_io_fini(env, io);
3308 cl_env_put(env, &refcheck);
3310 /* -ECANCELED indicates a matching lock with a different extent
3311 * was already present, and -EEXIST indicates a matching lock
3312 * on exactly the same extent was already present.
3313 * We convert them to positive values for userspace to make
3314 * recognizing true errors easier.
3315 * Note we can only return these detailed results on async requests,
3316 * as sync requests look the same as i/o requests for locking. */
3317 if (result == -ECANCELED)
3318 result = LLA_RESULT_DIFFERENT;
3319 else if (result == -EEXIST)
3320 result = LLA_RESULT_SAME;
3325 static const char *const ladvise_names[] = LU_LADVISE_NAMES;
3327 static int ll_ladvise_sanity(struct inode *inode,
3328 struct llapi_lu_ladvise *ladvise)
3330 struct ll_sb_info *sbi = ll_i2sbi(inode);
3331 enum lu_ladvise_type advice = ladvise->lla_advice;
3332 /* Note the peradvice flags is a 32 bit field, so per advice flags must
3333 * be in the first 32 bits of enum ladvise_flags */
3334 __u32 flags = ladvise->lla_peradvice_flags;
3335 /* 3 lines at 80 characters per line, should be plenty */
3338 if (advice > LU_LADVISE_MAX || advice == LU_LADVISE_INVALID) {
3341 "%s: advice with value '%d' not recognized, last supported advice is %s (value '%d'): rc = %d\n",
3342 sbi->ll_fsname, advice,
3343 ladvise_names[LU_LADVISE_MAX-1], LU_LADVISE_MAX-1, rc);
3347 /* Per-advice checks */
3349 case LU_LADVISE_LOCKNOEXPAND:
3350 if (flags & ~LF_LOCKNOEXPAND_MASK) {
3352 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3353 "rc = %d\n", sbi->ll_fsname, flags,
3354 ladvise_names[advice], rc);
3358 case LU_LADVISE_LOCKAHEAD:
3359 /* Currently only READ and WRITE modes can be requested */
3360 if (ladvise->lla_lockahead_mode >= MODE_MAX_USER ||
3361 ladvise->lla_lockahead_mode == 0) {
3363 CDEBUG(D_VFSTRACE, "%s: Invalid mode (%d) for %s: "
3364 "rc = %d\n", sbi->ll_fsname,
3365 ladvise->lla_lockahead_mode,
3366 ladvise_names[advice], rc);
3370 case LU_LADVISE_WILLREAD:
3371 case LU_LADVISE_DONTNEED:
3373 /* Note fall through above - These checks apply to all advices
3374 * except LOCKNOEXPAND */
3375 if (flags & ~LF_DEFAULT_MASK) {
3377 CDEBUG(D_VFSTRACE, "%s: Invalid flags (%x) for %s: "
3378 "rc = %d\n", sbi->ll_fsname, flags,
3379 ladvise_names[advice], rc);
3382 if (ladvise->lla_start >= ladvise->lla_end) {
3384 CDEBUG(D_VFSTRACE, "%s: Invalid range (%llu to %llu) "
3385 "for %s: rc = %d\n", sbi->ll_fsname,
3386 ladvise->lla_start, ladvise->lla_end,
3387 ladvise_names[advice], rc);
3399 * Give file access advices
3401 * The ladvise interface is similar to Linux fadvise() system call, except it
3402 * forwards the advices directly from Lustre client to server. The server side
3403 * codes will apply appropriate read-ahead and caching techniques for the
3404 * corresponding files.
3406 * A typical workload for ladvise is e.g. a bunch of different clients are
3407 * doing small random reads of a file, so prefetching pages into OSS cache
3408 * with big linear reads before the random IO is a net benefit. Fetching
3409 * all that data into each client cache with fadvise() may not be, due to
3410 * much more data being sent to the client.
3412 static int ll_ladvise(struct inode *inode, struct file *file, __u64 flags,
3413 struct llapi_lu_ladvise *ladvise)
3417 struct cl_ladvise_io *lio;
3422 env = cl_env_get(&refcheck);
3424 RETURN(PTR_ERR(env));
3426 io = vvp_env_thread_io(env);
3427 io->ci_obj = ll_i2info(inode)->lli_clob;
3429 /* initialize parameters for ladvise */
3430 lio = &io->u.ci_ladvise;
3431 lio->li_start = ladvise->lla_start;
3432 lio->li_end = ladvise->lla_end;
3433 lio->li_fid = ll_inode2fid(inode);
3434 lio->li_advice = ladvise->lla_advice;
3435 lio->li_flags = flags;
3437 if (cl_io_init(env, io, CIT_LADVISE, io->ci_obj) == 0)
3438 rc = cl_io_loop(env, io);
3442 cl_io_fini(env, io);
3443 cl_env_put(env, &refcheck);
3447 static int ll_lock_noexpand(struct file *file, int flags)
3449 struct ll_file_data *fd = file->private_data;
3451 fd->ll_lock_no_expand = !(flags & LF_UNSET);
3456 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
3459 struct fsxattr fsxattr;
3461 if (copy_from_user(&fsxattr,
3462 (const struct fsxattr __user *)arg,
3466 fsxattr.fsx_xflags = ll_inode_flags_to_xflags(inode->i_flags);
3467 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags))
3468 fsxattr.fsx_xflags |= FS_XFLAG_PROJINHERIT;
3469 fsxattr.fsx_projid = ll_i2info(inode)->lli_projid;
3470 if (copy_to_user((struct fsxattr __user *)arg,
3471 &fsxattr, sizeof(fsxattr)))
3477 int ll_ioctl_check_project(struct inode *inode, __u32 xflags,
3481 * Project Quota ID state is only allowed to change from within the init
3482 * namespace. Enforce that restriction only if we are trying to change
3483 * the quota ID state. Everything else is allowed in user namespaces.
3485 if (current_user_ns() == &init_user_ns) {
3487 * Caller is allowed to change the project ID. if it is being
3488 * changed, make sure that the new value is valid.
3490 if (ll_i2info(inode)->lli_projid != projid &&
3491 !projid_valid(make_kprojid(&init_user_ns, projid)))
3497 if (ll_i2info(inode)->lli_projid != projid)
3500 if (test_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags)) {
3501 if (!(xflags & FS_XFLAG_PROJINHERIT))
3504 if (xflags & FS_XFLAG_PROJINHERIT)
3511 static int ll_set_project(struct inode *inode, __u32 xflags, __u32 projid)
3513 struct ptlrpc_request *req = NULL;
3514 struct md_op_data *op_data;
3515 struct cl_object *obj;
3516 unsigned int inode_flags;
3519 rc = ll_ioctl_check_project(inode, xflags, projid);
3523 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
3524 LUSTRE_OPC_ANY, NULL);
3525 if (IS_ERR(op_data))
3526 RETURN(PTR_ERR(op_data));
3528 inode_flags = ll_xflags_to_inode_flags(xflags);
3529 op_data->op_attr_flags = ll_inode_to_ext_flags(inode_flags);
3530 if (xflags & FS_XFLAG_PROJINHERIT)
3531 op_data->op_attr_flags |= LUSTRE_PROJINHERIT_FL;
3533 /* pass projid to md_op_data */
3534 op_data->op_projid = projid;
3536 op_data->op_xvalid |= OP_XVALID_PROJID | OP_XVALID_FLAGS;
3537 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, NULL, 0, &req);
3538 ptlrpc_req_finished(req);
3540 GOTO(out_fsxattr, rc);
3541 ll_update_inode_flags(inode, op_data->op_attr_flags);
3543 /* Avoid OST RPC if this is only ioctl setting project inherit flag */
3544 if (xflags == 0 || xflags == FS_XFLAG_PROJINHERIT)
3545 GOTO(out_fsxattr, rc);
3547 obj = ll_i2info(inode)->lli_clob;
3549 struct iattr attr = { 0 };
3551 rc = cl_setattr_ost(obj, &attr, OP_XVALID_FLAGS, xflags);
3555 ll_finish_md_op_data(op_data);
3559 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
3562 struct fsxattr fsxattr;
3566 if (copy_from_user(&fsxattr,
3567 (const struct fsxattr __user *)arg,
3571 RETURN(ll_set_project(inode, fsxattr.fsx_xflags,
3572 fsxattr.fsx_projid));
3575 int ll_ioctl_project(struct file *file, unsigned int cmd,
3578 struct lu_project lu_project;
3579 struct dentry *dentry = file_dentry(file);
3580 struct inode *inode = file_inode(file);
3581 struct dentry *child_dentry = NULL;
3582 int rc = 0, name_len;
3584 if (copy_from_user(&lu_project,
3585 (const struct lu_project __user *)arg,
3586 sizeof(lu_project)))
3589 /* apply child dentry if name is valid */
3590 name_len = strnlen(lu_project.project_name, NAME_MAX);
3591 if (name_len > 0 && name_len <= NAME_MAX) {
3593 child_dentry = lookup_one_len(lu_project.project_name,
3595 inode_unlock(inode);
3596 if (IS_ERR(child_dentry)) {
3597 rc = PTR_ERR(child_dentry);
3600 inode = child_dentry->d_inode;
3605 } else if (name_len > NAME_MAX) {
3610 switch (lu_project.project_type) {
3611 case LU_PROJECT_SET:
3612 rc = ll_set_project(inode, lu_project.project_xflags,
3613 lu_project.project_id);
3615 case LU_PROJECT_GET:
3616 lu_project.project_xflags =
3617 ll_inode_flags_to_xflags(inode->i_flags);
3618 if (test_bit(LLIF_PROJECT_INHERIT,
3619 &ll_i2info(inode)->lli_flags))
3620 lu_project.project_xflags |= FS_XFLAG_PROJINHERIT;
3621 lu_project.project_id = ll_i2info(inode)->lli_projid;
3622 if (copy_to_user((struct lu_project __user *)arg,
3623 &lu_project, sizeof(lu_project))) {
3633 if (!IS_ERR_OR_NULL(child_dentry))
3638 static long ll_file_unlock_lease(struct file *file, struct ll_ioc_lease *ioc,
3641 struct inode *inode = file_inode(file);
3642 struct ll_file_data *fd = file->private_data;
3643 struct ll_inode_info *lli = ll_i2info(inode);
3644 struct obd_client_handle *och = NULL;
3645 struct split_param sp;
3646 struct pcc_param param;
3647 bool lease_broken = false;
3649 enum mds_op_bias bias = 0;
3650 struct file *layout_file = NULL;
3652 size_t data_size = 0;
3653 bool attached = false;
3658 mutex_lock(&lli->lli_och_mutex);
3659 if (fd->fd_lease_och != NULL) {
3660 och = fd->fd_lease_och;
3661 fd->fd_lease_och = NULL;
3663 mutex_unlock(&lli->lli_och_mutex);
3668 fmode = och->och_flags;
3670 switch (ioc->lil_flags) {
3671 case LL_LEASE_RESYNC_DONE:
3672 if (ioc->lil_count > IOC_IDS_MAX)
3673 GOTO(out_lease_close, rc = -EINVAL);
3675 data_size = offsetof(typeof(*ioc), lil_ids[ioc->lil_count]);
3676 OBD_ALLOC(data, data_size);
3678 GOTO(out_lease_close, rc = -ENOMEM);
3680 if (copy_from_user(data, (void __user *)arg, data_size))
3681 GOTO(out_lease_close, rc = -EFAULT);
3683 bias = MDS_CLOSE_RESYNC_DONE;
3685 case LL_LEASE_LAYOUT_MERGE: {
3688 if (ioc->lil_count != 1)
3689 GOTO(out_lease_close, rc = -EINVAL);
3691 arg += sizeof(*ioc);
3692 if (copy_from_user(&fd, (void __user *)arg, sizeof(__u32)))
3693 GOTO(out_lease_close, rc = -EFAULT);
3695 layout_file = fget(fd);
3697 GOTO(out_lease_close, rc = -EBADF);
3699 if ((file->f_flags & O_ACCMODE) == O_RDONLY ||
3700 (layout_file->f_flags & O_ACCMODE) == O_RDONLY)
3701 GOTO(out_lease_close, rc = -EPERM);
3703 data = file_inode(layout_file);
3704 bias = MDS_CLOSE_LAYOUT_MERGE;
3707 case LL_LEASE_LAYOUT_SPLIT: {
3711 if (ioc->lil_count != 2)
3712 GOTO(out_lease_close, rc = -EINVAL);
3714 arg += sizeof(*ioc);
3715 if (copy_from_user(&fdv, (void __user *)arg, sizeof(__u32)))
3716 GOTO(out_lease_close, rc = -EFAULT);
3718 arg += sizeof(__u32);
3719 if (copy_from_user(&mirror_id, (void __user *)arg,
3721 GOTO(out_lease_close, rc = -EFAULT);
3723 layout_file = fget(fdv);
3725 GOTO(out_lease_close, rc = -EBADF);
3727 /* if layout_file == file, it means to destroy the mirror */
3728 sp.sp_inode = file_inode(layout_file);
3729 sp.sp_mirror_id = (__u16)mirror_id;
3731 bias = MDS_CLOSE_LAYOUT_SPLIT;
3734 case LL_LEASE_PCC_ATTACH:
3735 if (ioc->lil_count != 1)
3738 if (IS_ENCRYPTED(inode))
3739 RETURN(-EOPNOTSUPP);
3741 arg += sizeof(*ioc);
3742 if (copy_from_user(¶m.pa_archive_id, (void __user *)arg,
3744 GOTO(out_lease_close, rc2 = -EFAULT);
3746 rc2 = pcc_readwrite_attach(file, inode, param.pa_archive_id);
3748 GOTO(out_lease_close, rc2);
3751 /* Grab latest data version */
3752 rc2 = ll_data_version(inode, ¶m.pa_data_version,
3755 GOTO(out_lease_close, rc2);
3758 bias = MDS_PCC_ATTACH;
3761 /* without close intent */
3766 rc = ll_lease_close_intent(och, inode, &lease_broken, bias, data);
3770 rc = ll_lease_och_release(inode, file);
3779 switch (ioc->lil_flags) {
3780 case LL_LEASE_RESYNC_DONE:
3782 OBD_FREE(data, data_size);
3784 case LL_LEASE_LAYOUT_MERGE:
3785 case LL_LEASE_LAYOUT_SPLIT:
3789 ll_layout_refresh(inode, &fd->fd_layout_version);
3791 case LL_LEASE_PCC_ATTACH:
3794 rc = pcc_readwrite_attach_fini(file, inode,
3795 param.pa_layout_gen,
3802 rc = ll_lease_type_from_fmode(fmode);
3806 static long ll_file_set_lease(struct file *file, struct ll_ioc_lease *ioc,
3809 struct inode *inode = file_inode(file);
3810 struct ll_inode_info *lli = ll_i2info(inode);
3811 struct ll_file_data *fd = file->private_data;
3812 struct obd_client_handle *och = NULL;
3813 __u64 open_flags = 0;
3819 switch (ioc->lil_mode) {
3820 case LL_LEASE_WRLCK:
3821 if (!(file->f_mode & FMODE_WRITE))
3823 fmode = FMODE_WRITE;
3825 case LL_LEASE_RDLCK:
3826 if (!(file->f_mode & FMODE_READ))
3830 case LL_LEASE_UNLCK:
3831 RETURN(ll_file_unlock_lease(file, ioc, arg));
3836 CDEBUG(D_INODE, "Set lease with mode %u\n", fmode);
3838 /* apply for lease */
3839 if (ioc->lil_flags & LL_LEASE_RESYNC)
3840 open_flags = MDS_OPEN_RESYNC;
3841 och = ll_lease_open(inode, file, fmode, open_flags);
3843 RETURN(PTR_ERR(och));
3845 if (ioc->lil_flags & LL_LEASE_RESYNC) {
3846 rc = ll_lease_file_resync(och, inode, arg);
3848 ll_lease_close(och, inode, NULL);
3851 rc = ll_layout_refresh(inode, &fd->fd_layout_version);
3853 ll_lease_close(och, inode, NULL);
3859 mutex_lock(&lli->lli_och_mutex);
3860 if (fd->fd_lease_och == NULL) {
3861 fd->fd_lease_och = och;
3864 mutex_unlock(&lli->lli_och_mutex);
3866 /* impossible now that only excl is supported for now */
3867 ll_lease_close(och, inode, &lease_broken);
3873 static void ll_heat_get(struct inode *inode, struct lu_heat *heat)
3875 struct ll_inode_info *lli = ll_i2info(inode);
3876 struct ll_sb_info *sbi = ll_i2sbi(inode);
3877 __u64 now = ktime_get_real_seconds();
3880 spin_lock(&lli->lli_heat_lock);
3881 heat->lh_flags = lli->lli_heat_flags;
3882 for (i = 0; i < heat->lh_count; i++)
3883 heat->lh_heat[i] = obd_heat_get(&lli->lli_heat_instances[i],
3884 now, sbi->ll_heat_decay_weight,
3885 sbi->ll_heat_period_second);
3886 spin_unlock(&lli->lli_heat_lock);
3889 static int ll_heat_set(struct inode *inode, enum lu_heat_flag flags)
3891 struct ll_inode_info *lli = ll_i2info(inode);
3894 spin_lock(&lli->lli_heat_lock);
3895 if (flags & LU_HEAT_FLAG_CLEAR)
3896 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
3898 if (flags & LU_HEAT_FLAG_OFF)
3899 lli->lli_heat_flags |= LU_HEAT_FLAG_OFF;
3901 lli->lli_heat_flags &= ~LU_HEAT_FLAG_OFF;
3903 spin_unlock(&lli->lli_heat_lock);
3909 ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3911 struct inode *inode = file_inode(file);
3912 struct ll_file_data *fd = file->private_data;
3916 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%x\n",
3917 PFID(ll_inode2fid(inode)), inode, cmd);
3918 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
3920 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
3921 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
3925 case LL_IOC_GETFLAGS:
3926 /* Get the current value of the file flags */
3927 return put_user(fd->fd_flags, (int __user *)arg);
3928 case LL_IOC_SETFLAGS:
3929 case LL_IOC_CLRFLAGS:
3930 /* Set or clear specific file flags */
3931 /* XXX This probably needs checks to ensure the flags are
3932 * not abused, and to handle any flag side effects.
3934 if (get_user(flags, (int __user *) arg))
3937 if (cmd == LL_IOC_SETFLAGS) {
3938 if ((flags & LL_FILE_IGNORE_LOCK) &&
3939 !(file->f_flags & O_DIRECT)) {
3940 CERROR("%s: unable to disable locking on "
3941 "non-O_DIRECT file\n", current->comm);
3945 fd->fd_flags |= flags;
3947 fd->fd_flags &= ~flags;
3950 case LL_IOC_LOV_SETSTRIPE:
3951 case LL_IOC_LOV_SETSTRIPE_NEW:
3952 RETURN(ll_lov_setstripe(inode, file, (void __user *)arg));
3953 case LL_IOC_LOV_SETEA:
3954 RETURN(ll_lov_setea(inode, file, (void __user *)arg));
3955 case LL_IOC_LOV_SWAP_LAYOUTS: {
3957 struct lustre_swap_layouts lsl;
3959 if (copy_from_user(&lsl, (char __user *)arg,
3960 sizeof(struct lustre_swap_layouts)))
3963 if ((file->f_flags & O_ACCMODE) == O_RDONLY)
3966 file2 = fget(lsl.sl_fd);
3970 /* O_WRONLY or O_RDWR */
3971 if ((file2->f_flags & O_ACCMODE) == O_RDONLY)
3972 GOTO(out, rc = -EPERM);
3974 if (lsl.sl_flags & SWAP_LAYOUTS_CLOSE) {
3975 struct inode *inode2;
3976 struct ll_inode_info *lli;
3977 struct obd_client_handle *och = NULL;
3979 lli = ll_i2info(inode);
3980 mutex_lock(&lli->lli_och_mutex);
3981 if (fd->fd_lease_och != NULL) {
3982 och = fd->fd_lease_och;
3983 fd->fd_lease_och = NULL;
3985 mutex_unlock(&lli->lli_och_mutex);
3987 GOTO(out, rc = -ENOLCK);
3988 inode2 = file_inode(file2);
3989 rc = ll_swap_layouts_close(och, inode, inode2);
3991 rc = ll_swap_layouts(file, file2, &lsl);
3997 case LL_IOC_LOV_GETSTRIPE:
3998 case LL_IOC_LOV_GETSTRIPE_NEW:
3999 RETURN(ll_file_getstripe(inode, (void __user *)arg, 0));
4000 case FS_IOC_GETFLAGS:
4001 case FS_IOC_SETFLAGS:
4002 RETURN(ll_iocontrol(inode, file, cmd, arg));
4003 case FSFILT_IOC_GETVERSION:
4004 case FS_IOC_GETVERSION:
4005 RETURN(put_user(inode->i_generation, (int __user *)arg));
4006 /* We need to special case any other ioctls we want to handle,
4007 * to send them to the MDS/OST as appropriate and to properly
4008 * network encode the arg field. */
4009 case FS_IOC_SETVERSION:
4012 case LL_IOC_GROUP_LOCK:
4013 RETURN(ll_get_grouplock(inode, file, arg));
4014 case LL_IOC_GROUP_UNLOCK:
4015 RETURN(ll_put_grouplock(inode, file, arg));
4016 case IOC_OBD_STATFS:
4017 RETURN(ll_obd_statfs(inode, (void __user *)arg));
4019 case LL_IOC_FLUSHCTX:
4020 RETURN(ll_flush_ctx(inode));
4021 case LL_IOC_PATH2FID: {
4022 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
4023 sizeof(struct lu_fid)))
4028 case LL_IOC_GETPARENT:
4029 RETURN(ll_getparent(file, (struct getparent __user *)arg));
4031 case OBD_IOC_FID2PATH:
4032 RETURN(ll_fid2path(inode, (void __user *)arg));
4033 case LL_IOC_DATA_VERSION: {
4034 struct ioc_data_version idv;
4037 if (copy_from_user(&idv, (char __user *)arg, sizeof(idv)))
4040 idv.idv_flags &= LL_DV_RD_FLUSH | LL_DV_WR_FLUSH;
4041 rc = ll_ioc_data_version(inode, &idv);
4044 copy_to_user((char __user *)arg, &idv, sizeof(idv)))
4050 case LL_IOC_GET_MDTIDX: {
4053 mdtidx = ll_get_mdt_idx(inode);
4057 if (put_user((int)mdtidx, (int __user *)arg))
4062 case OBD_IOC_GETNAME_OLD:
4063 case OBD_IOC_GETDTNAME:
4064 case OBD_IOC_GETMDNAME:
4065 RETURN(ll_get_obd_name(inode, cmd, arg));
4066 case LL_IOC_HSM_STATE_GET: {
4067 struct md_op_data *op_data;
4068 struct hsm_user_state *hus;
4075 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4076 LUSTRE_OPC_ANY, hus);
4077 if (IS_ERR(op_data)) {
4079 RETURN(PTR_ERR(op_data));
4082 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4085 if (copy_to_user((void __user *)arg, hus, sizeof(*hus)))
4088 ll_finish_md_op_data(op_data);
4092 case LL_IOC_HSM_STATE_SET: {
4093 struct hsm_state_set *hss;
4100 if (copy_from_user(hss, (char __user *)arg, sizeof(*hss))) {
4105 rc = ll_hsm_state_set(inode, hss);
4110 case LL_IOC_HSM_ACTION: {
4111 struct md_op_data *op_data;
4112 struct hsm_current_action *hca;
4120 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4121 LUSTRE_OPC_ANY, hca);
4122 if (IS_ERR(op_data)) {
4124 RETURN(PTR_ERR(op_data));
4127 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), sizeof(*op_data),
4130 GOTO(skip_copy, rc);
4132 /* The hsm_current_action retreived from the server could
4133 * contain corrupt information. If it is incorrect data collect
4134 * debug information. We still send the data even if incorrect
4135 * to user land to handle.
4137 action = hsm_user_action2name(hca->hca_action);
4138 if (strcmp(action, "UNKNOWN") == 0 ||
4139 hca->hca_state > HPS_DONE) {
4141 "HSM current state %s action %s, offset = %llu, length %llu\n",
4142 hsm_progress_state2name(hca->hca_state), action,
4143 hca->hca_location.offset, hca->hca_location.length);
4146 if (copy_to_user((char __user *)arg, hca, sizeof(*hca)))
4149 ll_finish_md_op_data(op_data);
4153 case LL_IOC_SET_LEASE_OLD: {
4154 struct ll_ioc_lease ioc = { .lil_mode = (__u32)arg };
4156 RETURN(ll_file_set_lease(file, &ioc, 0));
4158 case LL_IOC_SET_LEASE: {
4159 struct ll_ioc_lease ioc;
4161 if (copy_from_user(&ioc, (void __user *)arg, sizeof(ioc)))
4164 RETURN(ll_file_set_lease(file, &ioc, arg));
4166 case LL_IOC_GET_LEASE: {
4167 struct ll_inode_info *lli = ll_i2info(inode);
4168 struct ldlm_lock *lock = NULL;
4171 mutex_lock(&lli->lli_och_mutex);
4172 if (fd->fd_lease_och != NULL) {
4173 struct obd_client_handle *och = fd->fd_lease_och;
4175 lock = ldlm_handle2lock(&och->och_lease_handle);
4177 lock_res_and_lock(lock);
4178 if (!ldlm_is_cancel(lock))
4179 fmode = och->och_flags;
4181 unlock_res_and_lock(lock);
4182 LDLM_LOCK_PUT(lock);
4185 mutex_unlock(&lli->lli_och_mutex);
4187 RETURN(ll_lease_type_from_fmode(fmode));
4189 case LL_IOC_HSM_IMPORT: {
4190 struct hsm_user_import *hui;
4196 if (copy_from_user(hui, (void __user *)arg, sizeof(*hui))) {
4201 rc = ll_hsm_import(inode, file, hui);
4206 case LL_IOC_FUTIMES_3: {
4207 struct ll_futimes_3 lfu;
4209 if (copy_from_user(&lfu,
4210 (const struct ll_futimes_3 __user *)arg,
4214 RETURN(ll_file_futimes_3(file, &lfu));
4216 case LL_IOC_LADVISE: {
4217 struct llapi_ladvise_hdr *k_ladvise_hdr;
4218 struct llapi_ladvise_hdr __user *u_ladvise_hdr;
4221 int alloc_size = sizeof(*k_ladvise_hdr);
4224 u_ladvise_hdr = (void __user *)arg;
4225 OBD_ALLOC_PTR(k_ladvise_hdr);
4226 if (k_ladvise_hdr == NULL)
4229 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4230 GOTO(out_ladvise, rc = -EFAULT);
4232 if (k_ladvise_hdr->lah_magic != LADVISE_MAGIC ||
4233 k_ladvise_hdr->lah_count < 1)
4234 GOTO(out_ladvise, rc = -EINVAL);
4236 num_advise = k_ladvise_hdr->lah_count;
4237 if (num_advise >= LAH_COUNT_MAX)
4238 GOTO(out_ladvise, rc = -EFBIG);
4240 OBD_FREE_PTR(k_ladvise_hdr);
4241 alloc_size = offsetof(typeof(*k_ladvise_hdr),
4242 lah_advise[num_advise]);
4243 OBD_ALLOC(k_ladvise_hdr, alloc_size);
4244 if (k_ladvise_hdr == NULL)
4248 * TODO: submit multiple advices to one server in a single RPC
4250 if (copy_from_user(k_ladvise_hdr, u_ladvise_hdr, alloc_size))
4251 GOTO(out_ladvise, rc = -EFAULT);
4253 for (i = 0; i < num_advise; i++) {
4254 struct llapi_lu_ladvise *k_ladvise =
4255 &k_ladvise_hdr->lah_advise[i];
4256 struct llapi_lu_ladvise __user *u_ladvise =
4257 &u_ladvise_hdr->lah_advise[i];
4259 rc = ll_ladvise_sanity(inode, k_ladvise);
4261 GOTO(out_ladvise, rc);
4263 switch (k_ladvise->lla_advice) {
4264 case LU_LADVISE_LOCKNOEXPAND:
4265 rc = ll_lock_noexpand(file,
4266 k_ladvise->lla_peradvice_flags);
4267 GOTO(out_ladvise, rc);
4268 case LU_LADVISE_LOCKAHEAD:
4270 rc = ll_file_lock_ahead(file, k_ladvise);
4273 GOTO(out_ladvise, rc);
4276 &u_ladvise->lla_lockahead_result))
4277 GOTO(out_ladvise, rc = -EFAULT);
4280 rc = ll_ladvise(inode, file,
4281 k_ladvise_hdr->lah_flags,
4284 GOTO(out_ladvise, rc);
4291 OBD_FREE(k_ladvise_hdr, alloc_size);
4294 case LL_IOC_FLR_SET_MIRROR: {
4295 /* mirror I/O must be direct to avoid polluting page cache
4297 if (!(file->f_flags & O_DIRECT))
4300 fd->fd_designated_mirror = (__u32)arg;
4303 case FS_IOC_FSGETXATTR:
4304 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
4305 case FS_IOC_FSSETXATTR:
4306 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
4307 case LL_IOC_PROJECT:
4308 RETURN(ll_ioctl_project(file, cmd, arg));
4310 RETURN(put_user(PAGE_SIZE, (int __user *)arg));
4311 case LL_IOC_HEAT_GET: {
4312 struct lu_heat uheat;
4313 struct lu_heat *heat;
4316 if (copy_from_user(&uheat, (void __user *)arg, sizeof(uheat)))
4319 if (uheat.lh_count > OBD_HEAT_COUNT)
4320 uheat.lh_count = OBD_HEAT_COUNT;
4322 size = offsetof(typeof(uheat), lh_heat[uheat.lh_count]);
4323 OBD_ALLOC(heat, size);
4327 heat->lh_count = uheat.lh_count;
4328 ll_heat_get(inode, heat);
4329 rc = copy_to_user((char __user *)arg, heat, size);
4330 OBD_FREE(heat, size);
4331 RETURN(rc ? -EFAULT : 0);
4333 case LL_IOC_HEAT_SET: {
4336 if (copy_from_user(&flags, (void __user *)arg, sizeof(flags)))
4339 rc = ll_heat_set(inode, flags);
4342 case LL_IOC_PCC_DETACH: {
4343 struct lu_pcc_detach *detach;
4345 OBD_ALLOC_PTR(detach);
4349 if (copy_from_user(detach,
4350 (const struct lu_pcc_detach __user *)arg,
4352 GOTO(out_detach_free, rc = -EFAULT);
4354 if (!S_ISREG(inode->i_mode))
4355 GOTO(out_detach_free, rc = -EINVAL);
4357 if (!inode_owner_or_capable(&init_user_ns, inode))
4358 GOTO(out_detach_free, rc = -EPERM);
4360 rc = pcc_ioctl_detach(inode, detach->pccd_opt);
4362 OBD_FREE_PTR(detach);
4365 case LL_IOC_PCC_STATE: {
4366 struct lu_pcc_state __user *ustate =
4367 (struct lu_pcc_state __user *)arg;
4368 struct lu_pcc_state *state;
4370 OBD_ALLOC_PTR(state);
4374 if (copy_from_user(state, ustate, sizeof(*state)))
4375 GOTO(out_state, rc = -EFAULT);
4377 rc = pcc_ioctl_state(file, inode, state);
4379 GOTO(out_state, rc);
4381 if (copy_to_user(ustate, state, sizeof(*state)))
4382 GOTO(out_state, rc = -EFAULT);
4385 OBD_FREE_PTR(state);
4388 #ifdef HAVE_LUSTRE_CRYPTO
4389 case LL_IOC_SET_ENCRYPTION_POLICY:
4390 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4392 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
4393 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
4394 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4396 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
4397 case LL_IOC_ADD_ENCRYPTION_KEY:
4398 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4400 return llcrypt_ioctl_add_key(file, (void __user *)arg);
4401 case LL_IOC_REMOVE_ENCRYPTION_KEY:
4402 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4404 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
4405 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
4406 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4408 return llcrypt_ioctl_remove_key_all_users(file,
4409 (void __user *)arg);
4410 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
4411 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
4413 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
4416 case LL_IOC_UNLOCK_FOREIGN: {
4417 struct dentry *dentry = file_dentry(file);
4419 /* if not a foreign symlink do nothing */
4420 if (ll_foreign_is_removable(dentry, true)) {
4422 "prevent unlink of non-foreign file ("DFID")\n",
4423 PFID(ll_inode2fid(inode)));
4424 RETURN(-EOPNOTSUPP);
4430 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
4431 (void __user *)arg));
4435 loff_t ll_lseek(struct file *file, loff_t offset, int whence)
4437 struct inode *inode = file_inode(file);
4440 struct cl_lseek_io *lsio;
4447 env = cl_env_get(&refcheck);
4449 RETURN(PTR_ERR(env));
4451 io = vvp_env_thread_io(env);
4452 io->ci_obj = ll_i2info(inode)->lli_clob;
4453 ll_io_set_mirror(io, file);
4455 lsio = &io->u.ci_lseek;
4456 lsio->ls_start = offset;
4457 lsio->ls_whence = whence;
4458 lsio->ls_result = -ENXIO;
4461 rc = cl_io_init(env, io, CIT_LSEEK, io->ci_obj);
4463 struct vvp_io *vio = vvp_env_io(env);
4465 vio->vui_fd = file->private_data;
4466 rc = cl_io_loop(env, io);
4470 retval = rc ? : lsio->ls_result;
4471 cl_io_fini(env, io);
4472 } while (unlikely(io->ci_need_restart));
4474 cl_env_put(env, &refcheck);
4476 /* Without the key, SEEK_HOLE return value has to be
4477 * rounded up to next LUSTRE_ENCRYPTION_UNIT_SIZE.
4479 if (llcrypt_require_key(inode) == -ENOKEY && whence == SEEK_HOLE)
4480 retval = round_up(retval, LUSTRE_ENCRYPTION_UNIT_SIZE);
4485 static loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
4487 struct inode *inode = file_inode(file);
4488 loff_t retval = offset, eof = 0;
4489 ktime_t kstart = ktime_get();
4493 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), to=%llu=%#llx(%d)\n",
4494 PFID(ll_inode2fid(inode)), inode, retval, retval,
4497 if (origin == SEEK_END) {
4498 retval = ll_glimpse_size(inode);
4501 eof = i_size_read(inode);
4504 if (origin == SEEK_HOLE || origin == SEEK_DATA) {
4508 /* flush local cache first if any */
4509 cl_sync_file_range(inode, offset, OBD_OBJECT_EOF,
4512 retval = ll_lseek(file, offset, origin);
4515 retval = vfs_setpos(file, retval, ll_file_maxbytes(inode));
4517 retval = generic_file_llseek_size(file, offset, origin,
4518 ll_file_maxbytes(inode), eof);
4521 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK,
4522 ktime_us_delta(ktime_get(), kstart));
4526 static int ll_flush(struct file *file, fl_owner_t id)
4528 struct inode *inode = file_inode(file);
4529 struct ll_inode_info *lli = ll_i2info(inode);
4530 struct ll_file_data *fd = file->private_data;
4533 LASSERT(!S_ISDIR(inode->i_mode));
4535 /* catch async errors that were recorded back when async writeback
4536 * failed for pages in this mapping. */
4537 rc = lli->lli_async_rc;
4538 lli->lli_async_rc = 0;
4539 if (lli->lli_clob != NULL) {
4540 err = lov_read_and_clear_async_rc(lli->lli_clob);
4545 /* The application has been told write failure already.
4546 * Do not report failure again. */
4547 if (fd->fd_write_failed)
4549 return rc ? -EIO : 0;
4553 * Called to make sure a portion of file has been written out.
4554 * if @mode is not CL_FSYNC_LOCAL, it will send OST_SYNC RPCs to OST.
4556 * Return how many pages have been written.
4558 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
4559 enum cl_fsync_mode mode, int ignore_layout)
4563 struct cl_fsync_io *fio;
4568 if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL &&
4569 mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL)
4572 env = cl_env_get(&refcheck);
4574 RETURN(PTR_ERR(env));
4576 io = vvp_env_thread_io(env);
4577 io->ci_obj = ll_i2info(inode)->lli_clob;
4578 io->ci_ignore_layout = ignore_layout;
4580 /* initialize parameters for sync */
4581 fio = &io->u.ci_fsync;
4582 fio->fi_start = start;
4584 fio->fi_fid = ll_inode2fid(inode);
4585 fio->fi_mode = mode;
4586 fio->fi_nr_written = 0;
4588 if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0)
4589 result = cl_io_loop(env, io);
4591 result = io->ci_result;
4593 result = fio->fi_nr_written;
4594 cl_io_fini(env, io);
4595 cl_env_put(env, &refcheck);
4601 * When dentry is provided (the 'else' case), file_dentry() may be
4602 * null and dentry must be used directly rather than pulled from
4603 * file_dentry() as is done otherwise.
4606 int ll_fsync(struct file *file, loff_t start, loff_t end, int datasync)
4608 struct dentry *dentry = file_dentry(file);
4609 struct inode *inode = dentry->d_inode;
4610 struct ll_inode_info *lli = ll_i2info(inode);
4611 struct ptlrpc_request *req;
4612 ktime_t kstart = ktime_get();
4618 "VFS Op:inode="DFID"(%p), start %lld, end %lld, datasync %d\n",
4619 PFID(ll_inode2fid(inode)), inode, start, end, datasync);
4621 /* fsync's caller has already called _fdata{sync,write}, we want
4622 * that IO to finish before calling the osc and mdc sync methods */
4623 rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
4625 /* catch async errors that were recorded back when async writeback
4626 * failed for pages in this mapping. */
4627 if (!S_ISDIR(inode->i_mode)) {
4628 err = lli->lli_async_rc;
4629 lli->lli_async_rc = 0;
4632 if (lli->lli_clob != NULL) {
4633 err = lov_read_and_clear_async_rc(lli->lli_clob);
4639 err = md_fsync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), &req);
4643 ptlrpc_req_finished(req);
4645 if (S_ISREG(inode->i_mode)) {
4646 struct ll_file_data *fd = file->private_data;
4649 /* Sync metadata on MDT first, and then sync the cached data
4652 err = pcc_fsync(file, start, end, datasync, &cached);
4654 err = cl_sync_file_range(inode, start, end,
4656 if (rc == 0 && err < 0)
4659 fd->fd_write_failed = true;
4661 fd->fd_write_failed = false;
4665 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC,
4666 ktime_us_delta(ktime_get(), kstart));
4671 ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
4673 struct inode *inode = file_inode(file);
4674 struct ll_sb_info *sbi = ll_i2sbi(inode);
4675 struct ldlm_enqueue_info einfo = {
4676 .ei_type = LDLM_FLOCK,
4677 .ei_cb_cp = ldlm_flock_completion_ast,
4678 .ei_cbdata = file_lock,
4680 struct md_op_data *op_data;
4681 struct lustre_handle lockh = { 0 };
4682 union ldlm_policy_data flock = { { 0 } };
4683 int fl_type = file_lock->fl_type;
4684 ktime_t kstart = ktime_get();
4690 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID" file_lock=%p\n",
4691 PFID(ll_inode2fid(inode)), file_lock);
4693 if (file_lock->fl_flags & FL_FLOCK) {
4694 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
4695 /* flocks are whole-file locks */
4696 flock.l_flock.end = OFFSET_MAX;
4697 /* For flocks owner is determined by the local file desctiptor*/
4698 flock.l_flock.owner = (unsigned long)file_lock->fl_file;
4699 } else if (file_lock->fl_flags & FL_POSIX) {
4700 flock.l_flock.owner = (unsigned long)file_lock->fl_owner;
4701 flock.l_flock.start = file_lock->fl_start;
4702 flock.l_flock.end = file_lock->fl_end;
4706 flock.l_flock.pid = file_lock->fl_pid;
4708 #if defined(HAVE_LM_COMPARE_OWNER) || defined(lm_compare_owner)
4709 /* Somewhat ugly workaround for svc lockd.
4710 * lockd installs custom fl_lmops->lm_compare_owner that checks
4711 * for the fl_owner to be the same (which it always is on local node
4712 * I guess between lockd processes) and then compares pid.
4713 * As such we assign pid to the owner field to make it all work,
4714 * conflict with normal locks is unlikely since pid space and
4715 * pointer space for current->files are not intersecting */
4716 if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner)
4717 flock.l_flock.owner = (unsigned long)file_lock->fl_pid;
4722 einfo.ei_mode = LCK_PR;
4725 /* An unlock request may or may not have any relation to
4726 * existing locks so we may not be able to pass a lock handle
4727 * via a normal ldlm_lock_cancel() request. The request may even
4728 * unlock a byte range in the middle of an existing lock. In
4729 * order to process an unlock request we need all of the same
4730 * information that is given with a normal read or write record
4731 * lock request. To avoid creating another ldlm unlock (cancel)
4732 * message we'll treat a LCK_NL flock request as an unlock. */
4733 einfo.ei_mode = LCK_NL;
4736 einfo.ei_mode = LCK_PW;
4739 CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", fl_type);
4754 flags = LDLM_FL_BLOCK_NOWAIT;
4760 flags = LDLM_FL_TEST_LOCK;
4763 CERROR("unknown fcntl lock command: %d\n", cmd);
4767 /* Save the old mode so that if the mode in the lock changes we
4768 * can decrement the appropriate reader or writer refcount. */
4769 file_lock->fl_type = einfo.ei_mode;
4771 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
4772 LUSTRE_OPC_ANY, NULL);
4773 if (IS_ERR(op_data))
4774 RETURN(PTR_ERR(op_data));
4776 CDEBUG(D_DLMTRACE, "inode="DFID", pid=%u, flags=%#llx, mode=%u, "
4777 "start=%llu, end=%llu\n", PFID(ll_inode2fid(inode)),
4778 flock.l_flock.pid, flags, einfo.ei_mode,
4779 flock.l_flock.start, flock.l_flock.end);
4781 rc = md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data, &lockh,
4784 /* Restore the file lock type if not TEST lock. */
4785 if (!(flags & LDLM_FL_TEST_LOCK))
4786 file_lock->fl_type = fl_type;
4788 #ifdef HAVE_LOCKS_LOCK_FILE_WAIT
4789 if ((rc == 0 || file_lock->fl_type == F_UNLCK) &&
4790 !(flags & LDLM_FL_TEST_LOCK))
4791 rc2 = locks_lock_file_wait(file, file_lock);
4793 if ((file_lock->fl_flags & FL_FLOCK) &&
4794 (rc == 0 || file_lock->fl_type == F_UNLCK))
4795 rc2 = flock_lock_file_wait(file, file_lock);
4796 if ((file_lock->fl_flags & FL_POSIX) &&
4797 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
4798 !(flags & LDLM_FL_TEST_LOCK))
4799 rc2 = posix_lock_file_wait(file, file_lock);
4800 #endif /* HAVE_LOCKS_LOCK_FILE_WAIT */
4802 if (rc2 && file_lock->fl_type != F_UNLCK) {
4803 einfo.ei_mode = LCK_NL;
4804 md_enqueue(sbi->ll_md_exp, &einfo, &flock, op_data,
4809 ll_finish_md_op_data(op_data);
4812 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK,
4813 ktime_us_delta(ktime_get(), kstart));
4817 int ll_get_fid_by_name(struct inode *parent, const char *name,
4818 int namelen, struct lu_fid *fid,
4819 struct inode **inode)
4821 struct md_op_data *op_data = NULL;
4822 struct mdt_body *body;
4823 struct ptlrpc_request *req;
4827 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen, 0,
4828 LUSTRE_OPC_ANY, NULL);
4829 if (IS_ERR(op_data))
4830 RETURN(PTR_ERR(op_data));
4832 op_data->op_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
4833 rc = md_getattr_name(ll_i2sbi(parent)->ll_md_exp, op_data, &req);
4834 ll_finish_md_op_data(op_data);
4838 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
4840 GOTO(out_req, rc = -EFAULT);
4842 *fid = body->mbo_fid1;
4845 rc = ll_prep_inode(inode, &req->rq_pill, parent->i_sb, NULL);
4847 ptlrpc_req_finished(req);
4851 int ll_migrate(struct inode *parent, struct file *file, struct lmv_user_md *lum,
4852 const char *name, __u32 flags)
4854 struct dentry *dchild = NULL;
4855 struct inode *child_inode = NULL;
4856 struct md_op_data *op_data;
4857 struct ptlrpc_request *request = NULL;
4858 struct obd_client_handle *och = NULL;
4860 struct mdt_body *body;
4861 __u64 data_version = 0;
4862 size_t namelen = strlen(name);
4863 int lumlen = lmv_user_md_size(lum->lum_stripe_count, lum->lum_magic);
4867 CDEBUG(D_VFSTRACE, "migrate "DFID"/%s to MDT%04x stripe count %d\n",
4868 PFID(ll_inode2fid(parent)), name,
4869 lum->lum_stripe_offset, lum->lum_stripe_count);
4871 if (lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC) &&
4872 lum->lum_magic != cpu_to_le32(LMV_USER_MAGIC_SPECIFIC))
4873 lustre_swab_lmv_user_md(lum);
4875 /* Get child FID first */
4876 qstr.hash = ll_full_name_hash(file_dentry(file), name, namelen);
4879 dchild = d_lookup(file_dentry(file), &qstr);
4881 if (dchild->d_inode)
4882 child_inode = igrab(dchild->d_inode);
4887 rc = ll_get_fid_by_name(parent, name, namelen, NULL,
4896 if (!(exp_connect_flags2(ll_i2sbi(parent)->ll_md_exp) &
4897 OBD_CONNECT2_DIR_MIGRATE)) {
4898 if (le32_to_cpu(lum->lum_stripe_count) > 1 ||
4899 ll_dir_striped(child_inode)) {
4900 CERROR("%s: MDT doesn't support stripe directory "
4901 "migration!\n", ll_i2sbi(parent)->ll_fsname);
4902 GOTO(out_iput, rc = -EOPNOTSUPP);
4907 * lfs migrate command needs to be blocked on the client
4908 * by checking the migrate FID against the FID of the
4911 if (is_root_inode(child_inode))
4912 GOTO(out_iput, rc = -EINVAL);
4914 op_data = ll_prep_md_op_data(NULL, parent, NULL, name, namelen,
4915 child_inode->i_mode, LUSTRE_OPC_ANY, NULL);
4916 if (IS_ERR(op_data))
4917 GOTO(out_iput, rc = PTR_ERR(op_data));
4919 inode_lock(child_inode);
4920 op_data->op_fid3 = *ll_inode2fid(child_inode);
4921 if (!fid_is_sane(&op_data->op_fid3)) {
4922 CERROR("%s: migrate %s, but FID "DFID" is insane\n",
4923 ll_i2sbi(parent)->ll_fsname, name,
4924 PFID(&op_data->op_fid3));
4925 GOTO(out_unlock, rc = -EINVAL);
4928 op_data->op_cli_flags |= CLI_MIGRATE | CLI_SET_MEA;
4929 op_data->op_data = lum;
4930 op_data->op_data_size = lumlen;
4932 /* migrate dirent only for subdirs if MDS_MIGRATE_NSONLY set */
4933 if (S_ISDIR(child_inode->i_mode) && (flags & MDS_MIGRATE_NSONLY) &&
4934 lmv_dir_layout_changing(ll_i2info(parent)->lli_lsm_md))
4935 op_data->op_bias |= MDS_MIGRATE_NSONLY;
4938 if (S_ISREG(child_inode->i_mode)) {
4939 och = ll_lease_open(child_inode, NULL, FMODE_WRITE, 0);
4943 GOTO(out_unlock, rc);
4946 rc = ll_data_version(child_inode, &data_version,
4949 GOTO(out_close, rc);
4951 op_data->op_open_handle = och->och_open_handle;
4952 op_data->op_data_version = data_version;
4953 op_data->op_lease_handle = och->och_lease_handle;
4954 op_data->op_bias |= MDS_CLOSE_MIGRATE;
4956 spin_lock(&och->och_mod->mod_open_req->rq_lock);
4957 och->och_mod->mod_open_req->rq_replay = 0;
4958 spin_unlock(&och->och_mod->mod_open_req->rq_lock);
4961 rc = md_rename(ll_i2sbi(parent)->ll_md_exp, op_data,
4962 op_data->op_name, op_data->op_namelen,
4963 op_data->op_name, op_data->op_namelen, &request);
4965 LASSERT(request != NULL);
4966 ll_update_times(request, parent);
4969 if (rc == 0 || rc == -EAGAIN) {
4970 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
4971 LASSERT(body != NULL);
4973 /* If the server does release layout lock, then we cleanup
4974 * the client och here, otherwise release it in out_close: */
4975 if (och && body->mbo_valid & OBD_MD_CLOSE_INTENT_EXECED) {
4976 obd_mod_put(och->och_mod);
4977 md_clear_open_replay_data(ll_i2sbi(parent)->ll_md_exp,
4979 och->och_open_handle.cookie = DEAD_HANDLE_MAGIC;
4985 if (request != NULL) {
4986 ptlrpc_req_finished(request);
4990 /* Try again if the lease has cancelled. */
4991 if (rc == -EAGAIN && S_ISREG(child_inode->i_mode))
4996 ll_lease_close(och, child_inode, NULL);
4998 clear_nlink(child_inode);
5000 inode_unlock(child_inode);
5001 ll_finish_md_op_data(op_data);
5008 ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
5010 struct ll_file_data *fd = file->private_data;
5014 * In order to avoid flood of warning messages, only print one message
5015 * for one file. And the entire message rate on the client is limited
5016 * by CDEBUG_LIMIT too.
5018 if (!(fd->fd_flags & LL_FILE_FLOCK_WARNING)) {
5019 fd->fd_flags |= LL_FILE_FLOCK_WARNING;
5020 CDEBUG_LIMIT(D_CONSOLE,
5021 "flock disabled, mount with '-o [local]flock' to enable\r\n");
5027 * test if some locks matching bits and l_req_mode are acquired
5028 * - bits can be in different locks
5029 * - if found clear the common lock bits in *bits
5030 * - the bits not found, are kept in *bits
5032 * \param bits [IN] searched lock bits [IN]
5033 * \param l_req_mode [IN] searched lock mode
5034 * \retval boolean, true iff all bits are found
5036 int ll_have_md_lock(struct inode *inode, __u64 *bits, enum ldlm_mode l_req_mode)
5038 struct lustre_handle lockh;
5039 union ldlm_policy_data policy;
5040 enum ldlm_mode mode = (l_req_mode == LCK_MINMODE) ?
5041 (LCK_CR | LCK_CW | LCK_PR | LCK_PW) : l_req_mode;
5050 fid = &ll_i2info(inode)->lli_fid;
5051 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
5052 ldlm_lockname[mode]);
5054 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
5055 for (i = 0; i < MDS_INODELOCK_NUMBITS && *bits != 0; i++) {
5056 policy.l_inodebits.bits = *bits & BIT(i);
5057 if (policy.l_inodebits.bits == 0)
5060 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS,
5061 &policy, mode, &lockh)) {
5062 struct ldlm_lock *lock;
5064 lock = ldlm_handle2lock(&lockh);
5067 ~(lock->l_policy_data.l_inodebits.bits);
5068 LDLM_LOCK_PUT(lock);
5070 *bits &= ~policy.l_inodebits.bits;
5077 enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
5078 struct lustre_handle *lockh, __u64 flags,
5079 enum ldlm_mode mode)
5081 union ldlm_policy_data policy = { .l_inodebits = { bits } };
5086 fid = &ll_i2info(inode)->lli_fid;
5087 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
5089 rc = md_lock_match(ll_i2mdexp(inode), LDLM_FL_BLOCK_GRANTED|flags,
5090 fid, LDLM_IBITS, &policy, mode, lockh);
5095 static int ll_inode_revalidate_fini(struct inode *inode, int rc)
5097 /* Already unlinked. Just update nlink and return success */
5098 if (rc == -ENOENT) {
5100 /* If it is striped directory, and there is bad stripe
5101 * Let's revalidate the dentry again, instead of returning
5103 if (ll_dir_striped(inode))
5106 /* This path cannot be hit for regular files unless in
5107 * case of obscure races, so no need to to validate
5109 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
5111 } else if (rc != 0) {
5112 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
5113 "%s: revalidate FID "DFID" error: rc = %d\n",
5114 ll_i2sbi(inode)->ll_fsname,
5115 PFID(ll_inode2fid(inode)), rc);
5121 static int ll_inode_revalidate(struct dentry *dentry, enum ldlm_intent_flags op)
5123 struct inode *parent;
5124 struct inode *inode = dentry->d_inode;
5125 struct obd_export *exp = ll_i2mdexp(inode);
5126 struct lookup_intent oit = {
5129 struct ptlrpc_request *req = NULL;
5130 struct md_op_data *op_data;
5131 const char *name = NULL;
5136 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p),name=%s\n",
5137 PFID(ll_inode2fid(inode)), inode, dentry->d_name.name);
5139 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID) {
5140 parent = dentry->d_parent->d_inode;
5141 name = dentry->d_name.name;
5142 namelen = dentry->d_name.len;
5147 op_data = ll_prep_md_op_data(NULL, parent, inode, name, namelen, 0,
5148 LUSTRE_OPC_ANY, NULL);
5149 if (IS_ERR(op_data))
5150 RETURN(PTR_ERR(op_data));
5152 /* Call getattr by fid */
5153 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
5154 op_data->op_flags = MF_GETATTR_BY_FID;
5155 rc = md_intent_lock(exp, op_data, &oit, &req, &ll_md_blocking_ast, 0);
5156 ll_finish_md_op_data(op_data);
5158 rc = ll_inode_revalidate_fini(inode, rc);
5162 rc = ll_revalidate_it_finish(req, &oit, dentry);
5164 ll_intent_release(&oit);
5168 /* Unlinked? Unhash dentry, so it is not picked up later by
5169 * do_lookup() -> ll_revalidate_it(). We cannot use d_drop
5170 * here to preserve get_cwd functionality on 2.6.
5172 if (!dentry->d_inode->i_nlink)
5173 d_lustre_invalidate(dentry);
5175 ll_lookup_finish_locks(&oit, dentry);
5177 ptlrpc_req_finished(req);
5182 static int ll_merge_md_attr(struct inode *inode)
5184 struct ll_inode_info *lli = ll_i2info(inode);
5185 struct cl_attr attr = { 0 };
5188 if (!lli->lli_lsm_md)
5191 down_read(&lli->lli_lsm_sem);
5192 if (!lmv_dir_striped(lli->lli_lsm_md)) {
5193 up_read(&lli->lli_lsm_sem);
5196 rc = md_merge_attr(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
5197 &attr, ll_md_blocking_ast);
5198 up_read(&lli->lli_lsm_sem);
5202 spin_lock(&inode->i_lock);
5203 set_nlink(inode, attr.cat_nlink);
5204 spin_unlock(&inode->i_lock);
5206 inode->i_blocks = attr.cat_blocks;
5207 i_size_write(inode, attr.cat_size);
5209 ll_i2info(inode)->lli_atime = attr.cat_atime;
5210 ll_i2info(inode)->lli_mtime = attr.cat_mtime;
5211 ll_i2info(inode)->lli_ctime = attr.cat_ctime;
5216 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
5217 unsigned int flags, bool foreign)
5219 struct inode *inode = de->d_inode;
5220 struct ll_sb_info *sbi = ll_i2sbi(inode);
5221 struct ll_inode_info *lli = ll_i2info(inode);
5222 struct inode *dir = de->d_parent->d_inode;
5223 bool need_glimpse = true;
5224 ktime_t kstart = ktime_get();
5227 /* The OST object(s) determine the file size, blocks and mtime. */
5228 if (!(request_mask & STATX_SIZE || request_mask & STATX_BLOCKS ||
5229 request_mask & STATX_MTIME))
5230 need_glimpse = false;
5232 if (dentry_may_statahead(dir, de))
5233 ll_start_statahead(dir, de, need_glimpse &&
5234 !(flags & AT_STATX_DONT_SYNC));
5236 if (flags & AT_STATX_DONT_SYNC)
5237 GOTO(fill_attr, rc = 0);
5239 rc = ll_inode_revalidate(de, IT_GETATTR);
5243 /* foreign file/dir are always of zero length, so don't
5244 * need to validate size.
5246 if (S_ISREG(inode->i_mode) && !foreign) {
5250 GOTO(fill_attr, rc);
5252 rc = pcc_inode_getattr(inode, request_mask, flags, &cached);
5253 if (cached && rc < 0)
5257 GOTO(fill_attr, rc);
5260 * If the returned attr is masked with OBD_MD_FLSIZE &
5261 * OBD_MD_FLBLOCKS & OBD_MD_FLMTIME, it means that the file size
5262 * or blocks obtained from MDT is strictly correct, and the file
5263 * is usually not being modified by clients, and the [a|m|c]time
5264 * got from MDT is also strictly correct.
5265 * Under this circumstance, it does not need to send glimpse
5266 * RPCs to OSTs for file attributes such as the size and blocks.
5268 if (lli->lli_attr_valid & OBD_MD_FLSIZE &&
5269 lli->lli_attr_valid & OBD_MD_FLBLOCKS &&
5270 lli->lli_attr_valid & OBD_MD_FLMTIME) {
5271 inode->i_mtime.tv_sec = lli->lli_mtime;
5272 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5273 inode->i_atime.tv_sec = lli->lli_atime;
5274 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5275 inode->i_ctime.tv_sec = lli->lli_ctime;
5276 GOTO(fill_attr, rc);
5279 /* In case of restore, the MDT has the right size and has
5280 * already send it back without granting the layout lock,
5281 * inode is up-to-date so glimpse is useless.
5282 * Also to glimpse we need the layout, in case of a running
5283 * restore the MDT holds the layout lock so the glimpse will
5284 * block up to the end of restore (getattr will block)
5286 if (!test_bit(LLIF_FILE_RESTORING, &lli->lli_flags)) {
5287 rc = ll_glimpse_size(inode);
5292 /* If object isn't regular a file then don't validate size. */
5293 /* foreign dir is not striped dir */
5294 if (ll_dir_striped(inode) && !foreign) {
5295 rc = ll_merge_md_attr(inode);
5300 if (lli->lli_attr_valid & OBD_MD_FLATIME)
5301 inode->i_atime.tv_sec = lli->lli_atime;
5302 if (lli->lli_attr_valid & OBD_MD_FLMTIME)
5303 inode->i_mtime.tv_sec = lli->lli_mtime;
5304 if (lli->lli_attr_valid & OBD_MD_FLCTIME)
5305 inode->i_ctime.tv_sec = lli->lli_ctime;
5309 OBD_FAIL_TIMEOUT(OBD_FAIL_GETATTR_DELAY, 30);
5311 if (ll_need_32bit_api(sbi)) {
5312 stat->ino = cl_fid_build_ino(&lli->lli_fid, 1);
5313 stat->dev = ll_compat_encode_dev(inode->i_sb->s_dev);
5314 stat->rdev = ll_compat_encode_dev(inode->i_rdev);
5316 stat->ino = inode->i_ino;
5317 stat->dev = inode->i_sb->s_dev;
5318 stat->rdev = inode->i_rdev;
5321 /* foreign symlink to be exposed as a real symlink */
5323 stat->mode = inode->i_mode;
5325 stat->mode = (inode->i_mode & ~S_IFMT) | S_IFLNK;
5327 stat->uid = inode->i_uid;
5328 stat->gid = inode->i_gid;
5329 stat->atime = inode->i_atime;
5330 stat->mtime = inode->i_mtime;
5331 stat->ctime = inode->i_ctime;
5332 /* stat->blksize is used to tell about preferred IO size */
5333 if (sbi->ll_stat_blksize)
5334 stat->blksize = sbi->ll_stat_blksize;
5335 else if (S_ISREG(inode->i_mode))
5336 stat->blksize = 1 << min(PTLRPC_MAX_BRW_BITS + 1,
5337 LL_MAX_BLKSIZE_BITS);
5339 stat->blksize = 1 << inode->i_sb->s_blocksize_bits;
5341 stat->nlink = inode->i_nlink;
5342 stat->size = i_size_read(inode);
5343 stat->blocks = inode->i_blocks;
5345 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5346 if (flags & AT_STATX_DONT_SYNC) {
5347 if (stat->size == 0 &&
5348 lli->lli_attr_valid & OBD_MD_FLLAZYSIZE)
5349 stat->size = lli->lli_lazysize;
5350 if (stat->blocks == 0 &&
5351 lli->lli_attr_valid & OBD_MD_FLLAZYBLOCKS)
5352 stat->blocks = lli->lli_lazyblocks;
5355 if (lli->lli_attr_valid & OBD_MD_FLBTIME) {
5356 stat->result_mask |= STATX_BTIME;
5357 stat->btime.tv_sec = lli->lli_btime;
5360 stat->attributes_mask = STATX_ATTR_IMMUTABLE | STATX_ATTR_APPEND;
5361 stat->attributes |= ll_inode_to_ext_flags(inode->i_flags);
5362 stat->result_mask &= request_mask;
5365 ll_stats_ops_tally(sbi, LPROC_LL_GETATTR,
5366 ktime_us_delta(ktime_get(), kstart));
5371 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
5372 int ll_getattr(struct user_namespace *mnt_userns, const struct path *path,
5373 struct kstat *stat, u32 request_mask, unsigned int flags)
5375 return ll_getattr_dentry(path->dentry, stat, request_mask, flags,
5379 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
5381 return ll_getattr_dentry(de, stat, STATX_BASIC_STATS,
5382 AT_STATX_SYNC_AS_STAT, false);
5386 int cl_falloc(struct file *file, struct inode *inode, int mode, loff_t offset,
5389 loff_t size = i_size_read(inode);
5397 env = cl_env_get(&refcheck);
5399 RETURN(PTR_ERR(env));
5401 io = vvp_env_thread_io(env);
5402 io->ci_obj = ll_i2info(inode)->lli_clob;
5403 ll_io_set_mirror(io, file);
5405 io->ci_verify_layout = 1;
5406 io->u.ci_setattr.sa_parent_fid = lu_object_fid(&io->ci_obj->co_lu);
5407 io->u.ci_setattr.sa_falloc_mode = mode;
5408 io->u.ci_setattr.sa_falloc_offset = offset;
5409 io->u.ci_setattr.sa_falloc_end = offset + len;
5410 io->u.ci_setattr.sa_subtype = CL_SETATTR_FALLOCATE;
5412 CDEBUG(D_INODE, "UID %u GID %u PRJID %u\n",
5413 from_kuid(&init_user_ns, inode->i_uid),
5414 from_kgid(&init_user_ns, inode->i_gid),
5415 ll_i2info(inode)->lli_projid);
5417 io->u.ci_setattr.sa_falloc_uid = from_kuid(&init_user_ns, inode->i_uid);
5418 io->u.ci_setattr.sa_falloc_gid = from_kgid(&init_user_ns, inode->i_gid);
5419 io->u.ci_setattr.sa_falloc_projid = ll_i2info(inode)->lli_projid;
5421 if (io->u.ci_setattr.sa_falloc_end > size) {
5422 loff_t newsize = io->u.ci_setattr.sa_falloc_end;
5424 /* Check new size against VFS/VM file size limit and rlimit */
5425 rc = inode_newsize_ok(inode, newsize);
5428 if (newsize > ll_file_maxbytes(inode)) {
5429 CDEBUG(D_INODE, "file size too large %llu > %llu\n",
5430 (unsigned long long)newsize,
5431 ll_file_maxbytes(inode));
5438 rc = cl_io_init(env, io, CIT_SETATTR, io->ci_obj);
5440 rc = cl_io_loop(env, io);
5443 cl_io_fini(env, io);
5444 } while (unlikely(io->ci_need_restart));
5447 cl_env_put(env, &refcheck);
5451 long ll_fallocate(struct file *filp, int mode, loff_t offset, loff_t len)
5453 struct inode *inode = file_inode(filp);
5456 if (offset < 0 || len <= 0)
5459 * Encrypted inodes can't handle collapse range or zero range or insert
5460 * range since we would need to re-encrypt blocks with a different IV or
5461 * XTS tweak (which are based on the logical block number).
5462 * Similar to what ext4 does.
5464 if (IS_ENCRYPTED(inode) &&
5465 (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE |
5466 FALLOC_FL_ZERO_RANGE)))
5467 RETURN(-EOPNOTSUPP);
5470 * mode == 0 (which is standard prealloc) and PUNCH is supported
5471 * Rest of mode options are not supported yet.
5473 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
5474 RETURN(-EOPNOTSUPP);
5476 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FALLOCATE, 1);
5478 rc = cl_falloc(filp, inode, mode, offset, len);
5480 * ENOTSUPP (524) is an NFSv3 specific error code erroneously
5481 * used by Lustre in several places. Retuning it here would
5482 * confuse applications that explicity test for EOPNOTSUPP
5483 * (95) and fall back to ftruncate().
5485 if (rc == -ENOTSUPP)
5491 static int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5492 __u64 start, __u64 len)
5496 struct fiemap *fiemap;
5497 unsigned int extent_count = fieinfo->fi_extents_max;
5499 num_bytes = sizeof(*fiemap) + (extent_count *
5500 sizeof(struct fiemap_extent));
5501 OBD_ALLOC_LARGE(fiemap, num_bytes);
5506 fiemap->fm_flags = fieinfo->fi_flags;
5507 fiemap->fm_extent_count = fieinfo->fi_extents_max;
5508 fiemap->fm_start = start;
5509 fiemap->fm_length = len;
5510 if (extent_count > 0 &&
5511 copy_from_user(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
5512 sizeof(struct fiemap_extent)) != 0)
5513 GOTO(out, rc = -EFAULT);
5515 rc = ll_do_fiemap(inode, fiemap, num_bytes);
5517 if (IS_ENCRYPTED(inode)) {
5520 for (i = 0; i < fiemap->fm_mapped_extents; i++)
5521 fiemap->fm_extents[i].fe_flags |=
5522 FIEMAP_EXTENT_DATA_ENCRYPTED |
5523 FIEMAP_EXTENT_ENCODED;
5526 fieinfo->fi_flags = fiemap->fm_flags;
5527 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
5528 if (extent_count > 0 &&
5529 copy_to_user(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
5530 fiemap->fm_mapped_extents *
5531 sizeof(struct fiemap_extent)) != 0)
5532 GOTO(out, rc = -EFAULT);
5534 OBD_FREE_LARGE(fiemap, num_bytes);
5538 int ll_inode_permission(struct user_namespace *mnt_userns, struct inode *inode,
5542 struct ll_sb_info *sbi;
5543 struct root_squash_info *squash;
5544 struct cred *cred = NULL;
5545 const struct cred *old_cred = NULL;
5546 bool squash_id = false;
5547 ktime_t kstart = ktime_get();
5551 if (mask & MAY_NOT_BLOCK)
5555 * as root inode are NOT getting validated in lookup operation,
5556 * need to do it before permission check.
5559 if (is_root_inode(inode)) {
5560 rc = ll_inode_revalidate(inode->i_sb->s_root, IT_LOOKUP);
5565 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), inode mode %x mask %o\n",
5566 PFID(ll_inode2fid(inode)), inode, inode->i_mode, mask);
5568 /* squash fsuid/fsgid if needed */
5569 sbi = ll_i2sbi(inode);
5570 squash = &sbi->ll_squash;
5571 if (unlikely(squash->rsi_uid != 0 &&
5572 uid_eq(current_fsuid(), GLOBAL_ROOT_UID) &&
5573 !test_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags))) {
5577 CDEBUG(D_OTHER, "squash creds (%d:%d)=>(%d:%d)\n",
5578 __kuid_val(current_fsuid()), __kgid_val(current_fsgid()),
5579 squash->rsi_uid, squash->rsi_gid);
5581 /* update current process's credentials
5582 * and FS capability */
5583 cred = prepare_creds();
5587 cred->fsuid = make_kuid(&init_user_ns, squash->rsi_uid);
5588 cred->fsgid = make_kgid(&init_user_ns, squash->rsi_gid);
5589 cred->cap_effective = cap_drop_nfsd_set(cred->cap_effective);
5590 cred->cap_effective = cap_drop_fs_set(cred->cap_effective);
5592 old_cred = override_creds(cred);
5595 rc = generic_permission(mnt_userns, inode, mask);
5596 /* restore current process's credentials and FS capability */
5598 revert_creds(old_cred);
5603 ll_stats_ops_tally(sbi, LPROC_LL_INODE_PERM,
5604 ktime_us_delta(ktime_get(), kstart));
5609 /* -o localflock - only provides locally consistent flock locks */
5610 static const struct file_operations ll_file_operations = {
5611 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5612 # ifdef HAVE_SYNC_READ_WRITE
5613 .read = new_sync_read,
5614 .write = new_sync_write,
5616 .read_iter = ll_file_read_iter,
5617 .write_iter = ll_file_write_iter,
5618 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5619 .read = ll_file_read,
5620 .aio_read = ll_file_aio_read,
5621 .write = ll_file_write,
5622 .aio_write = ll_file_aio_write,
5623 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5624 .unlocked_ioctl = ll_file_ioctl,
5625 .open = ll_file_open,
5626 .release = ll_file_release,
5627 .mmap = ll_file_mmap,
5628 .llseek = ll_file_seek,
5629 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5630 .splice_read = generic_file_splice_read,
5632 .splice_read = pcc_file_splice_read,
5636 .fallocate = ll_fallocate,
5639 static const struct file_operations ll_file_operations_flock = {
5640 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5641 # ifdef HAVE_SYNC_READ_WRITE
5642 .read = new_sync_read,
5643 .write = new_sync_write,
5644 # endif /* HAVE_SYNC_READ_WRITE */
5645 .read_iter = ll_file_read_iter,
5646 .write_iter = ll_file_write_iter,
5647 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5648 .read = ll_file_read,
5649 .aio_read = ll_file_aio_read,
5650 .write = ll_file_write,
5651 .aio_write = ll_file_aio_write,
5652 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5653 .unlocked_ioctl = ll_file_ioctl,
5654 .open = ll_file_open,
5655 .release = ll_file_release,
5656 .mmap = ll_file_mmap,
5657 .llseek = ll_file_seek,
5658 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5659 .splice_read = generic_file_splice_read,
5661 .splice_read = pcc_file_splice_read,
5665 .flock = ll_file_flock,
5666 .lock = ll_file_flock,
5667 .fallocate = ll_fallocate,
5670 /* These are for -o noflock - to return ENOSYS on flock calls */
5671 static const struct file_operations ll_file_operations_noflock = {
5672 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
5673 # ifdef HAVE_SYNC_READ_WRITE
5674 .read = new_sync_read,
5675 .write = new_sync_write,
5676 # endif /* HAVE_SYNC_READ_WRITE */
5677 .read_iter = ll_file_read_iter,
5678 .write_iter = ll_file_write_iter,
5679 #else /* !HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5680 .read = ll_file_read,
5681 .aio_read = ll_file_aio_read,
5682 .write = ll_file_write,
5683 .aio_write = ll_file_aio_write,
5684 #endif /* HAVE_FILE_OPERATIONS_READ_WRITE_ITER */
5685 .unlocked_ioctl = ll_file_ioctl,
5686 .open = ll_file_open,
5687 .release = ll_file_release,
5688 .mmap = ll_file_mmap,
5689 .llseek = ll_file_seek,
5690 #ifndef HAVE_DEFAULT_FILE_SPLICE_READ_EXPORT
5691 .splice_read = generic_file_splice_read,
5693 .splice_read = pcc_file_splice_read,
5697 .flock = ll_file_noflock,
5698 .lock = ll_file_noflock,
5699 .fallocate = ll_fallocate,
5702 const struct inode_operations ll_file_inode_operations = {
5703 .setattr = ll_setattr,
5704 .getattr = ll_getattr,
5705 .permission = ll_inode_permission,
5706 #ifdef HAVE_IOP_XATTR
5707 .setxattr = ll_setxattr,
5708 .getxattr = ll_getxattr,
5709 .removexattr = ll_removexattr,
5711 .listxattr = ll_listxattr,
5712 .fiemap = ll_fiemap,
5713 .get_acl = ll_get_acl,
5714 #ifdef HAVE_IOP_SET_ACL
5715 .set_acl = ll_set_acl,
5719 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi)
5721 const struct file_operations *fops = &ll_file_operations_noflock;
5723 if (test_bit(LL_SBI_FLOCK, sbi->ll_flags))
5724 fops = &ll_file_operations_flock;
5725 else if (test_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags))
5726 fops = &ll_file_operations;
5731 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf)
5733 struct ll_inode_info *lli = ll_i2info(inode);
5734 struct cl_object *obj = lli->lli_clob;
5743 env = cl_env_get(&refcheck);
5745 RETURN(PTR_ERR(env));
5747 rc = cl_conf_set(env, lli->lli_clob, conf);
5751 if (conf->coc_opc == OBJECT_CONF_SET) {
5752 struct ldlm_lock *lock = conf->coc_lock;
5753 struct cl_layout cl = {
5757 LASSERT(lock != NULL);
5758 LASSERT(ldlm_has_layout(lock));
5760 /* it can only be allowed to match after layout is
5761 * applied to inode otherwise false layout would be
5762 * seen. Applying layout shoud happen before dropping
5763 * the intent lock. */
5764 ldlm_lock_allow_match(lock);
5766 rc = cl_object_layout_get(env, obj, &cl);
5771 DFID": layout version change: %u -> %u\n",
5772 PFID(&lli->lli_fid), ll_layout_version_get(lli),
5774 ll_layout_version_set(lli, cl.cl_layout_gen);
5778 cl_env_put(env, &refcheck);
5780 RETURN(rc < 0 ? rc : 0);
5783 /* Fetch layout from MDT with getxattr request, if it's not ready yet */
5784 static int ll_layout_fetch(struct inode *inode, struct ldlm_lock *lock)
5787 struct ll_sb_info *sbi = ll_i2sbi(inode);
5788 struct ptlrpc_request *req;
5795 CDEBUG(D_INODE, DFID" LVB_READY=%d l_lvb_data=%p l_lvb_len=%d\n",
5796 PFID(ll_inode2fid(inode)), ldlm_is_lvb_ready(lock),
5797 lock->l_lvb_data, lock->l_lvb_len);
5799 if (lock->l_lvb_data != NULL)
5802 /* if layout lock was granted right away, the layout is returned
5803 * within DLM_LVB of dlm reply; otherwise if the lock was ever
5804 * blocked and then granted via completion ast, we have to fetch
5805 * layout here. Please note that we can't use the LVB buffer in
5806 * completion AST because it doesn't have a large enough buffer */
5807 rc = ll_get_default_mdsize(sbi, &lmmsize);
5811 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode), OBD_MD_FLXATTR,
5812 XATTR_NAME_LOV, lmmsize, &req);
5815 GOTO(out, rc = 0); /* empty layout */
5822 if (lmmsize == 0) /* empty layout */
5825 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA, lmmsize);
5827 GOTO(out, rc = -EFAULT);
5829 OBD_ALLOC_LARGE(lvbdata, lmmsize);
5830 if (lvbdata == NULL)
5831 GOTO(out, rc = -ENOMEM);
5833 memcpy(lvbdata, lmm, lmmsize);
5834 lock_res_and_lock(lock);
5835 if (unlikely(lock->l_lvb_data == NULL)) {
5836 lock->l_lvb_type = LVB_T_LAYOUT;
5837 lock->l_lvb_data = lvbdata;
5838 lock->l_lvb_len = lmmsize;
5841 unlock_res_and_lock(lock);
5844 OBD_FREE_LARGE(lvbdata, lmmsize);
5849 ptlrpc_req_finished(req);
5854 * Apply the layout to the inode. Layout lock is held and will be released
5857 static int ll_layout_lock_set(struct lustre_handle *lockh, enum ldlm_mode mode,
5858 struct inode *inode)
5860 struct ll_inode_info *lli = ll_i2info(inode);
5861 struct ll_sb_info *sbi = ll_i2sbi(inode);
5862 struct ldlm_lock *lock;
5863 struct cl_object_conf conf;
5866 bool wait_layout = false;
5869 LASSERT(lustre_handle_is_used(lockh));
5871 lock = ldlm_handle2lock(lockh);
5872 LASSERT(lock != NULL);
5874 if (!ldlm_has_layout(lock))
5875 GOTO(out, rc = -EAGAIN);
5877 LDLM_DEBUG(lock, "file "DFID"(%p) being reconfigured",
5878 PFID(&lli->lli_fid), inode);
5880 /* in case this is a caching lock and reinstate with new inode */
5881 md_set_lock_data(sbi->ll_md_exp, lockh, inode, NULL);
5883 lock_res_and_lock(lock);
5884 lvb_ready = ldlm_is_lvb_ready(lock);
5885 unlock_res_and_lock(lock);
5887 /* checking lvb_ready is racy but this is okay. The worst case is
5888 * that multi processes may configure the file on the same time. */
5892 rc = ll_layout_fetch(inode, lock);
5896 /* for layout lock, lmm is stored in lock's lvb.
5897 * lvb_data is immutable if the lock is held so it's safe to access it
5900 * set layout to file. Unlikely this will fail as old layout was
5901 * surely eliminated */
5902 memset(&conf, 0, sizeof conf);
5903 conf.coc_opc = OBJECT_CONF_SET;
5904 conf.coc_inode = inode;
5905 conf.coc_lock = lock;
5906 conf.u.coc_layout.lb_buf = lock->l_lvb_data;
5907 conf.u.coc_layout.lb_len = lock->l_lvb_len;
5908 rc = ll_layout_conf(inode, &conf);
5910 /* refresh layout failed, need to wait */
5911 wait_layout = rc == -EBUSY;
5914 LDLM_LOCK_PUT(lock);
5915 ldlm_lock_decref(lockh, mode);
5917 /* wait for IO to complete if it's still being used. */
5919 CDEBUG(D_INODE, "%s: "DFID"(%p) wait for layout reconf\n",
5920 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5922 memset(&conf, 0, sizeof conf);
5923 conf.coc_opc = OBJECT_CONF_WAIT;
5924 conf.coc_inode = inode;
5925 rc = ll_layout_conf(inode, &conf);
5929 CDEBUG(D_INODE, "%s file="DFID" waiting layout return: %d\n",
5930 sbi->ll_fsname, PFID(&lli->lli_fid), rc);
5936 * Issue layout intent RPC to MDS.
5937 * \param inode [in] file inode
5938 * \param intent [in] layout intent
5940 * \retval 0 on success
5941 * \retval < 0 error code
5943 static int ll_layout_intent(struct inode *inode, struct layout_intent *intent)
5945 struct ll_inode_info *lli = ll_i2info(inode);
5946 struct ll_sb_info *sbi = ll_i2sbi(inode);
5947 struct md_op_data *op_data;
5948 struct lookup_intent it;
5949 struct ptlrpc_request *req;
5953 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL,
5954 0, 0, LUSTRE_OPC_ANY, NULL);
5955 if (IS_ERR(op_data))
5956 RETURN(PTR_ERR(op_data));
5958 op_data->op_data = intent;
5959 op_data->op_data_size = sizeof(*intent);
5961 memset(&it, 0, sizeof(it));
5962 it.it_op = IT_LAYOUT;
5963 if (intent->li_opc == LAYOUT_INTENT_WRITE ||
5964 intent->li_opc == LAYOUT_INTENT_TRUNC)
5965 it.it_flags = FMODE_WRITE;
5967 LDLM_DEBUG_NOLOCK("%s: requeue layout lock for file "DFID"(%p)",
5968 sbi->ll_fsname, PFID(&lli->lli_fid), inode);
5970 rc = md_intent_lock(sbi->ll_md_exp, op_data, &it, &req,
5971 &ll_md_blocking_ast, 0);
5972 if (it.it_request != NULL)
5973 ptlrpc_req_finished(it.it_request);
5974 it.it_request = NULL;
5976 ll_finish_md_op_data(op_data);
5978 /* set lock data in case this is a new lock */
5980 ll_set_lock_data(sbi->ll_md_exp, inode, &it, NULL);
5982 ll_intent_drop_lock(&it);
5988 * This function checks if there exists a LAYOUT lock on the client side,
5989 * or enqueues it if it doesn't have one in cache.
5991 * This function will not hold layout lock so it may be revoked any time after
5992 * this function returns. Any operations depend on layout should be redone
5995 * This function should be called before lov_io_init() to get an uptodate
5996 * layout version, the caller should save the version number and after IO
5997 * is finished, this function should be called again to verify that layout
5998 * is not changed during IO time.
6000 int ll_layout_refresh(struct inode *inode, __u32 *gen)
6002 struct ll_inode_info *lli = ll_i2info(inode);
6003 struct ll_sb_info *sbi = ll_i2sbi(inode);
6004 struct lustre_handle lockh;
6005 struct layout_intent intent = {
6006 .li_opc = LAYOUT_INTENT_ACCESS,
6008 enum ldlm_mode mode;
6012 *gen = ll_layout_version_get(lli);
6013 if (!test_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags) ||
6014 *gen != CL_LAYOUT_GEN_NONE)
6018 LASSERT(fid_is_sane(ll_inode2fid(inode)));
6019 LASSERT(S_ISREG(inode->i_mode));
6021 /* take layout lock mutex to enqueue layout lock exclusively. */
6022 mutex_lock(&lli->lli_layout_mutex);
6025 /* mostly layout lock is caching on the local side, so try to
6026 * match it before grabbing layout lock mutex. */
6027 mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh, 0,
6028 LCK_CR | LCK_CW | LCK_PR |
6030 if (mode != 0) { /* hit cached lock */
6031 rc = ll_layout_lock_set(&lockh, mode, inode);
6037 rc = ll_layout_intent(inode, &intent);
6043 *gen = ll_layout_version_get(lli);
6044 mutex_unlock(&lli->lli_layout_mutex);
6050 * Issue layout intent RPC indicating where in a file an IO is about to write.
6052 * \param[in] inode file inode.
6053 * \param[in] ext write range with start offset of fille in bytes where
6054 * an IO is about to write, and exclusive end offset in
6057 * \retval 0 on success
6058 * \retval < 0 error code
6060 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
6061 struct lu_extent *ext)
6063 struct layout_intent intent = {
6065 .li_extent.e_start = ext->e_start,
6066 .li_extent.e_end = ext->e_end,
6071 rc = ll_layout_intent(inode, &intent);
6077 * This function send a restore request to the MDT
6079 int ll_layout_restore(struct inode *inode, loff_t offset, __u64 length)
6081 struct hsm_user_request *hur;
6085 len = sizeof(struct hsm_user_request) +
6086 sizeof(struct hsm_user_item);
6087 OBD_ALLOC(hur, len);
6091 hur->hur_request.hr_action = HUA_RESTORE;
6092 hur->hur_request.hr_archive_id = 0;
6093 hur->hur_request.hr_flags = 0;
6094 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
6095 sizeof(hur->hur_user_item[0].hui_fid));
6096 hur->hur_user_item[0].hui_extent.offset = offset;
6097 hur->hur_user_item[0].hui_extent.length = length;
6098 hur->hur_request.hr_itemcount = 1;
6099 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,