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 * Directory code for lustre client.
37 #include <linux/pagemap.h>
39 #include <linux/version.h>
40 #include <linux/security.h>
41 #include <linux/user_namespace.h>
42 #include <linux/uidgid.h>
43 #include <linux/uaccess.h>
44 #include <linux/buffer_head.h> // for wait_on_buffer
45 #include <linux/pagevec.h>
47 #define DEBUG_SUBSYSTEM S_LLITE
49 #include <obd_support.h>
50 #include <obd_class.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <uapi/linux/llcrypt.h>
53 #include <lustre_lib.h>
54 #include <lustre_dlm.h>
55 #include <lustre_fid.h>
56 #include <lustre_kernelcomm.h>
57 #include <lustre_swab.h>
58 #include <libcfs/libcfs_crypto.h>
60 #include "llite_internal.h"
63 * (new) readdir implementation overview.
65 * Original lustre readdir implementation cached exact copy of raw directory
66 * pages on the client. These pages were indexed in client page cache by
67 * logical offset in the directory file. This design, while very simple and
68 * intuitive had some inherent problems:
70 * . it implies that byte offset to the directory entry serves as a
71 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
72 * ext3/htree directory entries may move due to splits, and more
75 * . it is incompatible with the design of split directories for cmd3,
76 * that assumes that names are distributed across nodes based on their
77 * hash, and so readdir should be done in hash order.
79 * New readdir implementation does readdir in hash order, and uses hash of a
80 * file name as a telldir/seekdir cookie. This led to number of complications:
82 * . hash is not unique, so it cannot be used to index cached directory
83 * pages on the client (note, that it requires a whole pageful of hash
84 * collided entries to cause two pages to have identical hashes);
86 * . hash is not unique, so it cannot, strictly speaking, be used as an
87 * entry cookie. ext3/htree has the same problem and lustre implementation
88 * mimics their solution: seekdir(hash) positions directory at the first
89 * entry with the given hash.
95 * Client caches directory pages using hash of the first entry as an index. As
96 * noted above hash is not unique, so this solution doesn't work as is:
97 * special processing is needed for "page hash chains" (i.e., sequences of
98 * pages filled with entries all having the same hash value).
100 * First, such chains have to be detected. To this end, server returns to the
101 * client the hash of the first entry on the page next to one returned. When
102 * client detects that this hash is the same as hash of the first entry on the
103 * returned page, page hash collision has to be handled. Pages in the
104 * hash chain, except first one, are termed "overflow pages".
106 * Proposed (unimplimented) solution to index uniqueness problem is to
107 * not cache overflow pages. Instead, when page hash collision is
108 * detected, all overflow pages from emerging chain should be
109 * immediately requested from the server and placed in a special data
110 * structure. This data structure can be used by ll_readdir() to
111 * process entries from overflow pages. When readdir invocation
112 * finishes, overflow pages are discarded. If page hash collision chain
113 * weren't completely processed, next call to readdir will again detect
114 * page hash collision, again read overflow pages in, process next
115 * portion of entries and again discard the pages. This is not as
116 * wasteful as it looks, because, given reasonable hash, page hash
117 * collisions are extremely rare.
119 * 1. directory positioning
121 * When seekdir(hash) is called, original
132 * identification of and access to overflow pages
136 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
137 * a header lu_dirpage which describes the start/end hash, and whether this
138 * page is empty (contains no dir entry) or hash collide with next page.
139 * After client receives reply, several pages will be integrated into dir page
140 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
141 * lu_dirpage for this integrated page will be adjusted. See
142 * mdc_adjust_dirpages().
145 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
148 struct md_callback cb_op;
152 cb_op.md_blocking_ast = ll_md_blocking_ast;
153 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
160 void ll_release_page(struct inode *inode, struct page *page,
165 /* Always remove the page for striped dir, because the page is
166 * built from temporarily in LMV layer */
167 if (inode && ll_dir_striped(inode)) {
174 if (likely(page->mapping != NULL))
175 delete_from_page_cache(page);
181 #ifdef HAVE_DIR_CONTEXT
182 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
183 struct dir_context *ctx)
186 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
187 void *cookie, filldir_t filldir)
190 struct ll_sb_info *sbi = ll_i2sbi(inode);
192 bool is_api32 = ll_need_32bit_api(sbi);
193 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
196 struct llcrypt_str lltr = LLTR_INIT(NULL, 0);
200 if (IS_ENCRYPTED(inode)) {
201 rc = llcrypt_fname_alloc_buffer(inode, NAME_MAX, &lltr);
206 page = ll_get_dir_page(inode, op_data, pos);
208 while (rc == 0 && !done) {
209 struct lu_dirpage *dp;
210 struct lu_dirent *ent;
219 hash = MDS_DIR_END_OFF;
220 dp = page_address(page);
221 for (ent = lu_dirent_start(dp); ent != NULL && !done;
222 ent = lu_dirent_next(ent)) {
229 hash = le64_to_cpu(ent->lde_hash);
230 if (hash < pos) /* Skip until we find target hash */
233 namelen = le16_to_cpu(ent->lde_namelen);
234 if (namelen == 0) /* Skip dummy record */
237 if (is_api32 && is_hash64)
241 fid_le_to_cpu(&fid, &ent->lde_fid);
242 ino = cl_fid_build_ino(&fid, is_api32);
243 type = S_DT(lu_dirent_type_get(ent));
244 /* For ll_nfs_get_name_filldir(), it will try to access
245 * 'ent' through 'lde_name', so the parameter 'name'
246 * for 'filldir()' must be part of the 'ent'. */
247 #ifdef HAVE_DIR_CONTEXT
249 if (!IS_ENCRYPTED(inode)) {
250 done = !dir_emit(ctx, ent->lde_name, namelen,
253 /* Directory is encrypted */
254 int save_len = lltr.len;
255 struct llcrypt_str de_name =
256 LLTR_INIT(ent->lde_name, namelen);
258 rc = ll_fname_disk_to_usr(inode, 0, 0, &de_name,
266 done = !dir_emit(ctx, de_name.name, de_name.len,
270 /* HAVE_DIR_CONTEXT is defined from kernel 3.11, whereas
271 * IS_ENCRYPTED is brought by kernel 4.14.
272 * So there is no need to handle encryption case here.
274 done = filldir(cookie, ent->lde_name, namelen, lhash,
281 ll_release_page(inode, page, false);
285 next = le64_to_cpu(dp->ldp_hash_end);
287 if (pos == MDS_DIR_END_OFF) {
289 * End of directory reached.
292 ll_release_page(inode, page, false);
295 * Normal case: continue to the next
298 ll_release_page(inode, page,
299 le32_to_cpu(dp->ldp_flags) &
302 page = ll_get_dir_page(inode, op_data, pos);
305 #ifdef HAVE_DIR_CONTEXT
310 llcrypt_fname_free_buffer(&lltr);
314 #ifdef HAVE_DIR_CONTEXT
315 static int ll_iterate(struct file *filp, struct dir_context *ctx)
317 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
320 struct inode *inode = file_inode(filp);
321 struct ll_file_data *lfd = filp->private_data;
322 struct ll_sb_info *sbi = ll_i2sbi(inode);
323 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
324 int api32 = ll_need_32bit_api(sbi);
325 struct md_op_data *op_data;
326 struct lu_fid pfid = { 0 };
327 ktime_t kstart = ktime_get();
338 "VFS Op:inode="DFID"(%p) pos/size%lu/%llu 32bit_api %d\n",
339 PFID(ll_inode2fid(inode)),
340 inode, (unsigned long)pos, i_size_read(inode), api32);
342 if (IS_ENCRYPTED(inode)) {
343 rc = llcrypt_get_encryption_info(inode);
344 if (rc && rc != -ENOKEY)
348 if (pos == MDS_DIR_END_OFF)
354 if (unlikely(ll_dir_striped(inode))) {
356 * This is only needed for striped dir to fill ..,
357 * see lmv_read_page()
359 if (file_dentry(filp)->d_parent != NULL &&
360 file_dentry(filp)->d_parent->d_inode != NULL) {
361 __u64 ibits = MDS_INODELOCK_LOOKUP;
362 struct inode *parent =
363 file_dentry(filp)->d_parent->d_inode;
365 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
366 pfid = *ll_inode2fid(parent);
369 /* If it can not find in cache, do lookup .. on the master
371 if (fid_is_zero(&pfid)) {
372 rc = ll_dir_get_parent_fid(inode, &pfid);
378 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
379 LUSTRE_OPC_ANY, inode);
381 GOTO(out, rc = PTR_ERR(op_data));
383 /* foreign dirs are browsed out of Lustre */
384 if (unlikely(op_data->op_mea1 != NULL &&
385 op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
386 ll_finish_md_op_data(op_data);
390 op_data->op_fid3 = pfid;
392 #ifdef HAVE_DIR_CONTEXT
394 rc = ll_dir_read(inode, &pos, op_data, ctx);
397 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
402 if (pos == MDS_DIR_END_OFF) {
404 pos = LL_DIR_END_OFF_32BIT;
406 pos = LL_DIR_END_OFF;
411 #ifdef HAVE_DIR_CONTEXT
416 ll_finish_md_op_data(op_data);
420 ll_stats_ops_tally(sbi, LPROC_LL_READDIR,
421 ktime_us_delta(ktime_get(), kstart));
427 * Create striped directory with specified stripe(@lump)
429 * \param[in] dparent the parent of the directory.
430 * \param[in] lump the specified stripes.
431 * \param[in] dirname the name of the directory.
432 * \param[in] mode the specified mode of the directory.
434 * \retval =0 if striped directory is being created successfully.
435 * <0 if the creation is failed.
437 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
438 size_t len, const char *dirname, umode_t mode,
441 struct inode *parent = dparent->d_inode;
442 struct ptlrpc_request *request = NULL;
443 struct md_op_data *op_data;
444 struct ll_sb_info *sbi = ll_i2sbi(parent);
445 struct inode *inode = NULL;
446 struct dentry dentry = {
450 .len = strlen(dirname),
451 .hash = ll_full_name_hash(dparent, dirname,
455 bool encrypt = false;
459 if (unlikely(!lmv_user_magic_supported(lump->lum_magic)))
462 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
464 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
465 PFID(ll_inode2fid(parent)), parent, dirname,
466 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
468 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
471 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
472 PFID(ll_inode2fid(parent)), parent, dirname,
473 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
476 if (lump->lum_stripe_count > 1 &&
477 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
480 if (IS_DEADDIR(parent) &&
481 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
484 /* MDS < 2.14 doesn't support 'crush' hash type, and cannot handle
485 * unknown hash if client doesn't set a valid one. switch to fnv_1a_64.
487 if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_CRUSH)) {
488 enum lmv_hash_type type = lump->lum_hash_type &
491 if (type == LMV_HASH_TYPE_CRUSH ||
492 type == LMV_HASH_TYPE_UNKNOWN)
493 lump->lum_hash_type = (lump->lum_hash_type ^ type) |
494 LMV_HASH_TYPE_FNV_1A_64;
497 if (unlikely(!lmv_user_magic_supported(cpu_to_le32(lump->lum_magic))))
498 lustre_swab_lmv_user_md(lump);
500 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
501 mode &= ~current_umask();
502 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
503 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
504 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
507 RETURN(PTR_ERR(op_data));
509 op_data->op_dir_depth = ll_i2info(parent)->lli_depth;
511 if (ll_sbi_has_encrypt(sbi) &&
512 (IS_ENCRYPTED(parent) ||
513 unlikely(llcrypt_dummy_context_enabled(parent)))) {
514 err = llcrypt_get_encryption_info(parent);
516 GOTO(out_op_data, err);
517 if (!llcrypt_has_encryption_key(parent))
518 GOTO(out_op_data, err = -ENOKEY);
522 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
523 /* selinux_dentry_init_security() uses dentry->d_parent and name
524 * to determine the security context for the file. So our fake
525 * dentry should be real enough for this purpose. */
526 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
527 &op_data->op_file_secctx_name,
528 &op_data->op_file_secctx,
529 &op_data->op_file_secctx_size);
531 GOTO(out_op_data, err);
535 err = llcrypt_inherit_context(parent, NULL, op_data, false);
537 GOTO(out_op_data, err);
540 op_data->op_cli_flags |= CLI_SET_MEA;
542 op_data->op_bias |= MDS_SETSTRIPE_CREATE;
544 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
545 from_kuid(&init_user_ns, current_fsuid()),
546 from_kgid(&init_user_ns, current_fsgid()),
547 current_cap(), 0, &request);
549 GOTO(out_request, err);
551 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
553 err = ll_prep_inode(&inode, &request->rq_pill, parent->i_sb, NULL);
555 GOTO(out_inode, err);
557 dentry.d_inode = inode;
559 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
560 /* no need to protect selinux_inode_setsecurity() by
561 * inode_lock. Taking it would lead to a client deadlock
564 err = security_inode_notifysecctx(inode,
565 op_data->op_file_secctx,
566 op_data->op_file_secctx_size);
568 err = ll_inode_init_security(&dentry, inode, parent);
571 GOTO(out_inode, err);
574 err = ll_set_encflags(inode, op_data->op_file_encctx,
575 op_data->op_file_encctx_size, false);
577 GOTO(out_inode, err);
583 ptlrpc_req_finished(request);
585 ll_finish_md_op_data(op_data);
590 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
593 struct ll_sb_info *sbi = ll_i2sbi(inode);
594 struct md_op_data *op_data;
595 struct ptlrpc_request *req = NULL;
601 switch (lump->lmm_magic) {
602 case LOV_USER_MAGIC_V1:
603 lum_size = sizeof(struct lov_user_md_v1);
605 case LOV_USER_MAGIC_V3:
606 lum_size = sizeof(struct lov_user_md_v3);
608 case LOV_USER_MAGIC_COMP_V1:
609 lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
612 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
613 lustre_swab_lmv_user_md(
614 (struct lmv_user_md *)lump);
615 lum_size = sizeof(struct lmv_user_md);
617 case LOV_USER_MAGIC_SPECIFIC: {
618 struct lov_user_md_v3 *v3 =
619 (struct lov_user_md_v3 *)lump;
620 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
622 lum_size = lov_user_md_size(v3->lmm_stripe_count,
623 LOV_USER_MAGIC_SPECIFIC);
627 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
628 " %#08x != %#08x nor %#08x\n",
629 lump->lmm_magic, LOV_USER_MAGIC_V1,
635 * This is coming from userspace, so should be in
636 * local endian. But the MDS would like it in little
637 * endian, so we swab it before we send it.
639 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
640 le32_to_cpu(LOV_MAGIC_MAGIC))
641 lustre_swab_lov_user_md(lump, 0);
643 lum_size = sizeof(struct lov_user_md_v1);
646 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
647 LUSTRE_OPC_ANY, NULL);
649 RETURN(PTR_ERR(op_data));
651 /* swabbing is done in lov_setstripe() on server side */
652 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
653 ll_finish_md_op_data(op_data);
654 ptlrpc_req_finished(req);
661 static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
663 struct ptlrpc_request **request, u64 valid,
664 enum get_default_layout_type type)
666 struct ll_sb_info *sbi = ll_i2sbi(inode);
667 struct mdt_body *body;
668 struct lov_mds_md *lmm = NULL;
669 struct ptlrpc_request *req = NULL;
671 struct md_op_data *op_data;
675 rc = ll_get_default_mdsize(sbi, &lmm_size);
679 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
680 0, lmm_size, LUSTRE_OPC_ANY,
683 RETURN(PTR_ERR(op_data));
685 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
687 if (type == GET_DEFAULT_LAYOUT_ROOT) {
688 lu_root_fid(&op_data->op_fid1);
689 fid = op_data->op_fid1;
691 fid = *ll_inode2fid(inode);
694 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
695 ll_finish_md_op_data(op_data);
697 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
702 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
703 LASSERT(body != NULL);
705 lmm_size = body->mbo_eadatasize;
707 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
709 GOTO(out, rc = -ENODATA);
712 lmm = req_capsule_server_sized_get(&req->rq_pill,
713 &RMF_MDT_MD, lmm_size);
714 LASSERT(lmm != NULL);
717 * This is coming from the MDS, so is probably in
718 * little endian. We convert it to host endian before
719 * passing it to userspace.
721 /* We don't swab objects for directories */
722 switch (le32_to_cpu(lmm->lmm_magic)) {
725 case LOV_MAGIC_COMP_V1:
726 case LOV_USER_MAGIC_SPECIFIC:
727 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
728 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
731 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
732 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
735 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
736 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
738 case LMV_MAGIC_FOREIGN: {
739 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
741 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
742 __swab32s(&lfm->lfm_magic);
743 __swab32s(&lfm->lfm_length);
744 __swab32s(&lfm->lfm_type);
745 __swab32s(&lfm->lfm_flags);
750 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
755 *plmm_size = lmm_size;
761 * This function will be used to get default LOV/LMV/Default LMV
762 * @valid will be used to indicate which stripe it will retrieve.
763 * If the directory does not have its own default layout, then the
764 * function will request the default layout from root FID.
765 * OBD_MD_MEA LMV stripe EA
766 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
767 * otherwise Default LOV EA.
768 * Each time, it can only retrieve 1 stripe EA
770 int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
771 struct ptlrpc_request **request,
772 struct ptlrpc_request **root_request,
775 struct ptlrpc_request *req = NULL;
776 struct ptlrpc_request *root_req = NULL;
777 struct lov_mds_md *lmm = NULL;
782 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
784 if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
785 !(valid & (OBD_MD_MEA|OBD_MD_DEFAULT_MEA)) && root_request != NULL){
786 int rc2 = ll_dir_get_default_layout(inode, (void **)&lmm,
787 &lmm_size, &root_req, valid,
788 GET_DEFAULT_LAYOUT_ROOT);
794 *plmm_size = lmm_size;
796 if (root_request != NULL)
797 *root_request = root_req;
803 * This function will be used to get default LOV/LMV/Default LMV
804 * @valid will be used to indicate which stripe it will retrieve
805 * OBD_MD_MEA LMV stripe EA
806 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
807 * otherwise Default LOV EA.
808 * Each time, it can only retrieve 1 stripe EA
810 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
811 struct ptlrpc_request **request, u64 valid)
813 struct ptlrpc_request *req = NULL;
814 struct lov_mds_md *lmm = NULL;
819 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
823 *plmm_size = lmm_size;
829 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
831 struct md_op_data *op_data;
836 OBD_ALLOC_PTR(op_data);
840 op_data->op_flags |= MF_GET_MDT_IDX;
841 op_data->op_fid1 = *fid;
842 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
843 mdt_index = op_data->op_mds;
844 OBD_FREE_PTR(op_data);
852 * Get MDT index for the inode.
854 int ll_get_mdt_idx(struct inode *inode)
856 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
860 * Generic handler to do any pre-copy work.
862 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
863 * first information for it that real work has started.
865 * Moreover, for a ARCHIVE request, it will sample the file data version and
866 * store it in \a copy.
868 * \return 0 on success.
870 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
872 struct ll_sb_info *sbi = ll_s2sbi(sb);
873 struct hsm_progress_kernel hpk;
878 /* Forge a hsm_progress based on data from copy. */
879 hpk.hpk_fid = copy->hc_hai.hai_fid;
880 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
881 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
882 hpk.hpk_extent.length = 0;
885 hpk.hpk_data_version = 0;
888 /* For archive request, we need to read the current file version. */
889 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
891 __u64 data_version = 0;
893 /* Get inode for this fid */
894 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
896 hpk.hpk_flags |= HP_FLAG_RETRY;
897 /* hpk_errval is >= 0 */
898 hpk.hpk_errval = -PTR_ERR(inode);
899 GOTO(progress, rc = PTR_ERR(inode));
902 /* Read current file data version */
903 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
906 CDEBUG(D_HSM, "Could not read file data version of "
907 DFID" (rc = %d). Archive request ("
908 "%#llx) could not be done.\n",
909 PFID(©->hc_hai.hai_fid), rc,
910 copy->hc_hai.hai_cookie);
911 hpk.hpk_flags |= HP_FLAG_RETRY;
912 /* hpk_errval must be >= 0 */
913 hpk.hpk_errval = -rc;
917 /* Store in the hsm_copy for later copytool use.
918 * Always modified even if no lsm. */
919 copy->hc_data_version = data_version;
923 /* On error, the request should be considered as completed */
924 if (hpk.hpk_errval > 0)
925 hpk.hpk_flags |= HP_FLAG_COMPLETED;
927 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
930 /* Return first error */
931 RETURN(rc != 0 ? rc : rc2);
935 * Generic handler to do any post-copy work.
937 * It will send the last hsm_progress update to coordinator to inform it
938 * that copy is finished and whether it was successful or not.
941 * - for ARCHIVE request, it will sample the file data version and compare it
942 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
943 * will be considered as failed.
944 * - for RESTORE request, it will sample the file data version and send it to
945 * coordinator which is useful if the file was imported as 'released'.
947 * \return 0 on success.
949 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
951 struct ll_sb_info *sbi = ll_s2sbi(sb);
952 struct hsm_progress_kernel hpk;
957 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
958 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
959 * initialized if copy_end was called with copy == NULL.
962 /* Forge a hsm_progress based on data from copy. */
963 hpk.hpk_fid = copy->hc_hai.hai_fid;
964 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
965 hpk.hpk_extent = copy->hc_hai.hai_extent;
966 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
967 hpk.hpk_errval = copy->hc_errval;
968 hpk.hpk_data_version = 0;
970 /* For archive request, we need to check the file data was not changed.
972 * For restore request, we need to send the file data version, this is
973 * useful when the file was created using hsm_import.
975 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
976 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
977 (copy->hc_errval == 0)) {
979 __u64 data_version = 0;
981 /* Get lsm for this fid */
982 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
984 hpk.hpk_flags |= HP_FLAG_RETRY;
985 /* hpk_errval must be >= 0 */
986 hpk.hpk_errval = -PTR_ERR(inode);
987 GOTO(progress, rc = PTR_ERR(inode));
990 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
993 CDEBUG(D_HSM, "Could not read file data version. "
994 "Request could not be confirmed.\n");
995 if (hpk.hpk_errval == 0)
996 hpk.hpk_errval = -rc;
1000 /* Store in the hsm_copy for later copytool use.
1001 * Always modified even if no lsm. */
1002 hpk.hpk_data_version = data_version;
1004 /* File could have been stripped during archiving, so we need
1005 * to check anyway. */
1006 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1007 (copy->hc_data_version != data_version)) {
1008 CDEBUG(D_HSM, "File data version mismatched. "
1009 "File content was changed during archiving. "
1010 DFID", start:%#llx current:%#llx\n",
1011 PFID(©->hc_hai.hai_fid),
1012 copy->hc_data_version, data_version);
1013 /* File was changed, send error to cdt. Do not ask for
1014 * retry because if a file is modified frequently,
1015 * the cdt will loop on retried archive requests.
1016 * The policy engine will ask for a new archive later
1017 * when the file will not be modified for some tunable
1019 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1021 /* hpk_errval must be >= 0 */
1022 hpk.hpk_errval = -rc;
1029 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1032 /* Return first error */
1033 RETURN(rc != 0 ? rc : rc2);
1037 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1038 const struct lustre_kernelcomm __user *data)
1040 struct lustre_kernelcomm *lk;
1041 struct lustre_kernelcomm *tmp;
1042 size_t size = sizeof(*lk);
1047 /* copy data from userspace to get numbers of archive_id */
1048 OBD_ALLOC(lk, size);
1052 if (copy_from_user(lk, data, size))
1053 GOTO(out_lk, rc = -EFAULT);
1055 if (lk->lk_flags & LK_FLG_STOP)
1058 if (!(lk->lk_flags & LK_FLG_DATANR)) {
1059 __u32 archive_mask = lk->lk_data_count;
1062 /* old hsm agent to old MDS */
1063 if (!exp_connect_archive_id_array(exp))
1066 /* old hsm agent to new MDS */
1067 lk->lk_flags |= LK_FLG_DATANR;
1069 if (archive_mask == 0)
1072 count = hweight32(archive_mask);
1073 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1074 OBD_ALLOC(tmp, new_size);
1076 GOTO(out_lk, rc = -ENOMEM);
1078 memcpy(tmp, lk, size);
1079 tmp->lk_data_count = count;
1085 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1086 if (BIT(i) & archive_mask) {
1087 lk->lk_data[count] = i + 1;
1094 /* new hsm agent to new mds */
1095 if (lk->lk_data_count > 0) {
1096 new_size = offsetof(struct lustre_kernelcomm,
1097 lk_data[lk->lk_data_count]);
1098 OBD_ALLOC(tmp, new_size);
1100 GOTO(out_lk, rc = -ENOMEM);
1106 if (copy_from_user(lk, data, size))
1107 GOTO(out_lk, rc = -EFAULT);
1110 /* new hsm agent to old MDS */
1111 if (!exp_connect_archive_id_array(exp)) {
1114 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1115 GOTO(out_lk, rc = -EINVAL);
1117 for (i = 0; i < lk->lk_data_count; i++) {
1118 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1120 CERROR("%s: archive id %d requested but only "
1121 "[0 - %zu] supported: rc = %d\n",
1122 exp->exp_obd->obd_name, lk->lk_data[i],
1123 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1127 if (lk->lk_data[i] == 0) {
1132 archives |= (1 << (lk->lk_data[i] - 1));
1134 lk->lk_flags &= ~LK_FLG_DATANR;
1135 lk->lk_data_count = archives;
1138 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1144 static int check_owner(int type, int id)
1148 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1152 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1161 int quotactl_ioctl(struct super_block *sb, struct if_quotactl *qctl)
1163 struct ll_sb_info *sbi = ll_s2sbi(sb);
1164 int cmd = qctl->qc_cmd;
1165 int type = qctl->qc_type;
1166 int id = qctl->qc_id;
1167 int valid = qctl->qc_valid;
1175 case LUSTRE_Q_SETDEFAULT:
1176 case LUSTRE_Q_SETQUOTAPOOL:
1177 case LUSTRE_Q_SETINFOPOOL:
1178 case LUSTRE_Q_SETDEFAULT_POOL:
1179 if (!capable(CAP_SYS_ADMIN))
1182 if (sb->s_flags & SB_RDONLY)
1186 case LUSTRE_Q_GETDEFAULT:
1187 case LUSTRE_Q_GETQUOTAPOOL:
1188 case LUSTRE_Q_GETDEFAULT_POOL:
1189 if (check_owner(type, id) &&
1190 (!capable(CAP_SYS_ADMIN)))
1194 case LUSTRE_Q_GETINFOPOOL:
1197 CERROR("unsupported quotactl op: %#x\n", cmd);
1201 if (valid != QC_GENERAL) {
1202 if (cmd == Q_GETINFO)
1203 qctl->qc_cmd = Q_GETOINFO;
1204 else if (cmd == Q_GETQUOTA ||
1205 cmd == LUSTRE_Q_GETQUOTAPOOL)
1206 qctl->qc_cmd = Q_GETOQUOTA;
1212 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1213 sizeof(*qctl), qctl, NULL);
1216 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1217 sizeof(*qctl), qctl, NULL);
1220 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1221 sizeof(*qctl), qctl, NULL);
1223 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1225 sizeof(*qctl), qctl, NULL);
1237 struct obd_quotactl *oqctl;
1238 int oqctl_len = sizeof(*oqctl);
1240 if (LUSTRE_Q_CMD_IS_POOL(cmd))
1241 oqctl_len += LOV_MAXPOOLNAME + 1;
1243 OBD_ALLOC(oqctl, oqctl_len);
1247 QCTL_COPY(oqctl, qctl);
1248 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1250 OBD_FREE(oqctl, oqctl_len);
1253 /* If QIF_SPACE is not set, client should collect the
1254 * space usage from OSSs by itself */
1255 if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
1256 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1257 !oqctl->qc_dqblk.dqb_curspace) {
1258 struct obd_quotactl *oqctl_tmp;
1259 int qctl_len = sizeof(*oqctl_tmp) + LOV_MAXPOOLNAME + 1;
1261 OBD_ALLOC(oqctl_tmp, qctl_len);
1262 if (oqctl_tmp == NULL)
1263 GOTO(out, rc = -ENOMEM);
1265 if (cmd == LUSTRE_Q_GETQUOTAPOOL) {
1266 oqctl_tmp->qc_cmd = LUSTRE_Q_GETQUOTAPOOL;
1267 memcpy(oqctl_tmp->qc_poolname,
1269 LOV_MAXPOOLNAME + 1);
1271 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1273 oqctl_tmp->qc_id = oqctl->qc_id;
1274 oqctl_tmp->qc_type = oqctl->qc_type;
1276 /* collect space usage from OSTs */
1277 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1278 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1279 if (!rc || rc == -EREMOTEIO) {
1280 oqctl->qc_dqblk.dqb_curspace =
1281 oqctl_tmp->qc_dqblk.dqb_curspace;
1282 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1285 /* collect space & inode usage from MDTs */
1286 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1287 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1288 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1289 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1290 if (!rc || rc == -EREMOTEIO) {
1291 oqctl->qc_dqblk.dqb_curspace +=
1292 oqctl_tmp->qc_dqblk.dqb_curspace;
1293 oqctl->qc_dqblk.dqb_curinodes =
1294 oqctl_tmp->qc_dqblk.dqb_curinodes;
1295 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1297 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1300 OBD_FREE(oqctl_tmp, qctl_len);
1303 QCTL_COPY(qctl, oqctl);
1304 OBD_FREE(oqctl, oqctl_len);
1310 int ll_rmfid(struct file *file, void __user *arg)
1312 const struct fid_array __user *ufa = arg;
1313 struct fid_array *lfa = NULL;
1316 int i, rc, *rcs = NULL;
1319 if (!capable(CAP_DAC_READ_SEARCH) &&
1320 !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1322 /* Only need to get the buflen */
1323 if (get_user(nr, &ufa->fa_nr))
1325 /* DoS protection */
1326 if (nr > OBD_MAX_FIDS_IN_ARRAY)
1329 size = offsetof(struct fid_array, fa_fids[nr]);
1330 OBD_ALLOC(lfa, size);
1333 OBD_ALLOC_PTR_ARRAY(rcs, nr);
1335 GOTO(free_lfa, rc = -ENOMEM);
1337 if (copy_from_user(lfa, arg, size))
1338 GOTO(free_rcs, rc = -EFAULT);
1340 /* Call mdc_iocontrol */
1341 rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1343 for (i = 0; i < nr; i++)
1345 lfa->fa_fids[i].f_ver = rcs[i];
1346 if (copy_to_user(arg, lfa, size))
1351 OBD_FREE_PTR_ARRAY(rcs, nr);
1353 OBD_FREE(lfa, size);
1358 /* This function tries to get a single name component,
1359 * to send to the server. No actual path traversal involved,
1360 * so we limit to NAME_MAX */
1361 static char *ll_getname(const char __user *filename)
1366 OBD_ALLOC(tmp, NAME_MAX + 1);
1369 return ERR_PTR(-ENOMEM);
1371 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1374 else if (len > NAME_MAX)
1375 ret = -ENAMETOOLONG;
1378 OBD_FREE(tmp, NAME_MAX + 1);
1384 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1386 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1388 struct dentry *dentry = file_dentry(file);
1389 struct inode *inode = file_inode(file);
1390 struct ll_sb_info *sbi = ll_i2sbi(inode);
1391 struct obd_ioctl_data *data = NULL;
1395 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1396 PFID(ll_inode2fid(inode)), inode, cmd);
1398 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1399 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1402 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1404 case FS_IOC_GETFLAGS:
1405 case FS_IOC_SETFLAGS:
1406 RETURN(ll_iocontrol(inode, file, cmd, arg));
1407 case FSFILT_IOC_GETVERSION:
1408 case FS_IOC_GETVERSION:
1409 RETURN(put_user(inode->i_generation, (int __user *)arg));
1410 /* We need to special case any other ioctls we want to handle,
1411 * to send them to the MDS/OST as appropriate and to properly
1412 * network encode the arg field. */
1413 case FS_IOC_SETVERSION:
1416 case LL_IOC_GET_MDTIDX: {
1419 mdtidx = ll_get_mdt_idx(inode);
1423 if (put_user((int)mdtidx, (int __user *)arg))
1428 case IOC_MDC_LOOKUP: {
1429 int namelen, len = 0;
1432 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
1436 filename = data->ioc_inlbuf1;
1437 namelen = strlen(filename);
1439 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1440 GOTO(out_free, rc = -EINVAL);
1443 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1445 CERROR("%s: lookup %.*s failed: rc = %d\n",
1446 sbi->ll_fsname, namelen, filename, rc);
1450 OBD_FREE_LARGE(data, len);
1453 case LL_IOC_LMV_SETSTRIPE: {
1454 struct lmv_user_md *lum;
1459 bool createonly = false;
1463 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
1467 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1468 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1469 GOTO(lmv_out_free, rc = -EINVAL);
1471 filename = data->ioc_inlbuf1;
1472 namelen = data->ioc_inllen1;
1475 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1476 GOTO(lmv_out_free, rc = -EINVAL);
1478 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1479 lumlen = data->ioc_inllen2;
1481 if (!lmv_user_magic_supported(lum->lum_magic)) {
1482 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1483 lum->lum_magic, -EINVAL);
1484 GOTO(lmv_out_free, rc = -EINVAL);
1487 if ((lum->lum_magic == LMV_USER_MAGIC ||
1488 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1489 lumlen < sizeof(*lum)) {
1490 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1491 filename, lumlen, lum->lum_magic, -EINVAL);
1492 GOTO(lmv_out_free, rc = -EINVAL);
1495 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1496 lumlen < sizeof(struct lmv_foreign_md)) {
1497 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1498 filename, lum->lum_magic, lumlen, -EFAULT);
1499 GOTO(lmv_out_free, rc = -EINVAL);
1502 mode = data->ioc_type;
1503 createonly = data->ioc_obdo1.o_flags & OBD_FL_OBDMDEXISTS;
1504 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode,
1507 OBD_FREE_LARGE(data, len);
1511 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1512 struct lmv_user_md lum;
1513 struct lmv_user_md __user *ulump =
1514 (struct lmv_user_md __user *)arg;
1517 if (copy_from_user(&lum, ulump, sizeof(lum)))
1520 if (lum.lum_magic != LMV_USER_MAGIC)
1523 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1527 case LL_IOC_LOV_SETSTRIPE_NEW:
1528 case LL_IOC_LOV_SETSTRIPE: {
1529 struct lov_user_md_v3 *lumv3 = NULL;
1530 struct lov_user_md_v1 lumv1;
1531 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1532 struct lov_user_md_v1 __user *lumv1p =
1533 (struct lov_user_md_v1 __user *)arg;
1534 struct lov_user_md_v3 __user *lumv3p =
1535 (struct lov_user_md_v3 __user *)arg;
1538 int set_default = 0;
1540 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1541 sizeof(struct lov_comp_md_v1));
1542 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1543 /* first try with v1 which is smaller than v3 */
1544 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1547 if (is_root_inode(inode))
1550 switch (lumv1.lmm_magic) {
1551 case LOV_USER_MAGIC_V3:
1552 case LOV_USER_MAGIC_SPECIFIC:
1553 lum_size = ll_lov_user_md_size(&lumv1);
1556 OBD_ALLOC(lumv3, lum_size);
1559 if (copy_from_user(lumv3, lumv3p, lum_size))
1560 GOTO(out, rc = -EFAULT);
1561 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1563 case LOV_USER_MAGIC_V1:
1566 GOTO(out, rc = -ENOTSUPP);
1569 /* in v1 and v3 cases lumv1 points to data */
1570 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1573 OBD_FREE(lumv3, lum_size);
1576 case LL_IOC_LMV_GETSTRIPE: {
1577 struct lmv_user_md __user *ulmv =
1578 (struct lmv_user_md __user *)arg;
1579 struct lmv_user_md lum;
1580 struct ptlrpc_request *request = NULL;
1581 struct ptlrpc_request *root_request = NULL;
1582 union lmv_mds_md *lmm = NULL;
1585 struct lmv_user_md *tmp = NULL;
1589 int max_stripe_count;
1593 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1596 max_stripe_count = lum.lum_stripe_count;
1597 /* lum_magic will indicate which stripe the ioctl will like
1598 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1599 * is for default LMV stripe */
1600 if (lum.lum_magic == LMV_MAGIC_V1)
1601 valid |= OBD_MD_MEA;
1602 else if (lum.lum_magic == LMV_USER_MAGIC)
1603 valid |= OBD_MD_DEFAULT_MEA;
1607 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1608 &request, &root_request, valid);
1610 GOTO(finish_req, rc);
1612 /* Get default LMV EA */
1613 if (lum.lum_magic == LMV_USER_MAGIC) {
1614 if (lmmsize > sizeof(*ulmv))
1615 GOTO(finish_req, rc = -EINVAL);
1617 if (copy_to_user(ulmv, lmm, lmmsize))
1618 GOTO(finish_req, rc = -EFAULT);
1620 GOTO(finish_req, rc);
1623 /* if foreign LMV case, fake stripes number */
1624 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1625 struct lmv_foreign_md *lfm;
1627 lfm = (struct lmv_foreign_md *)lmm;
1628 if (lfm->lfm_length < XATTR_SIZE_MAX -
1629 offsetof(typeof(*lfm), lfm_value)) {
1630 __u32 size = lfm->lfm_length +
1631 offsetof(typeof(*lfm), lfm_value);
1633 stripe_count = lmv_foreign_to_md_stripes(size);
1635 CERROR("invalid %d foreign size returned\n",
1640 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1642 if (max_stripe_count < stripe_count) {
1643 lum.lum_stripe_count = stripe_count;
1644 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1645 GOTO(finish_req, rc = -EFAULT);
1646 GOTO(finish_req, rc = -E2BIG);
1649 /* enough room on user side and foreign case */
1650 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1651 struct lmv_foreign_md *lfm;
1654 lfm = (struct lmv_foreign_md *)lmm;
1655 size = lfm->lfm_length +
1656 offsetof(struct lmv_foreign_md, lfm_value);
1657 if (copy_to_user(ulmv, lfm, size))
1658 GOTO(finish_req, rc = -EFAULT);
1659 GOTO(finish_req, rc);
1662 lum_size = lmv_user_md_size(stripe_count,
1663 LMV_USER_MAGIC_SPECIFIC);
1664 OBD_ALLOC(tmp, lum_size);
1666 GOTO(finish_req, rc = -ENOMEM);
1668 mdt_index = ll_get_mdt_idx(inode);
1670 GOTO(out_tmp, rc = -ENOMEM);
1672 tmp->lum_magic = LMV_MAGIC_V1;
1673 tmp->lum_stripe_count = 0;
1674 tmp->lum_stripe_offset = mdt_index;
1675 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1676 for (i = 0; i < stripe_count; i++) {
1679 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1680 if (fid_is_sane(&fid)) {
1681 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1683 GOTO(out_tmp, rc = mdt_index);
1685 tmp->lum_objects[i].lum_mds = mdt_index;
1686 tmp->lum_objects[i].lum_fid = fid;
1689 tmp->lum_stripe_count++;
1692 if (copy_to_user(ulmv, tmp, lum_size))
1693 GOTO(out_tmp, rc = -EFAULT);
1695 OBD_FREE(tmp, lum_size);
1697 ptlrpc_req_finished(request);
1698 ptlrpc_req_finished(root_request);
1702 case LL_IOC_UNLOCK_FOREIGN:
1703 /* if not a foreign symlink do nothing */
1704 if (ll_foreign_is_removable(dentry, true)) {
1706 "prevent rmdir of non-foreign dir ("DFID")\n",
1707 PFID(ll_inode2fid(inode)));
1708 RETURN(-EOPNOTSUPP);
1712 case LL_IOC_REMOVE_ENTRY: {
1713 char *filename = NULL;
1717 /* Here is a little hack to avoid sending REINT_RMENTRY to
1718 * unsupported server, which might crash the server(LU-2730),
1719 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1720 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1721 * server will support REINT_RMENTRY XXX*/
1722 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1725 filename = ll_getname((const char __user *)arg);
1726 if (IS_ERR(filename))
1727 RETURN(PTR_ERR(filename));
1729 namelen = strlen(filename);
1731 GOTO(out_rmdir, rc = -EINVAL);
1733 rc = ll_rmdir_entry(inode, filename, namelen);
1736 ll_putname(filename);
1740 RETURN(ll_rmfid(file, (void __user *)arg));
1741 case LL_IOC_LOV_SWAP_LAYOUTS:
1743 case IOC_OBD_STATFS:
1744 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1745 case LL_IOC_LOV_GETSTRIPE:
1746 case LL_IOC_LOV_GETSTRIPE_NEW:
1747 case LL_IOC_MDC_GETINFO_V1:
1748 case LL_IOC_MDC_GETINFO_V2:
1749 case IOC_MDC_GETFILEINFO_V1:
1750 case IOC_MDC_GETFILEINFO_V2:
1751 case IOC_MDC_GETFILESTRIPE: {
1752 struct ptlrpc_request *request = NULL;
1753 struct ptlrpc_request *root_request = NULL;
1754 struct lov_user_md __user *lump;
1755 struct lov_mds_md *lmm = NULL;
1756 struct mdt_body *body;
1757 char *filename = NULL;
1758 lstat_t __user *statp = NULL;
1759 lstatx_t __user *stxp = NULL;
1760 __u64 __user *flagsp = NULL;
1761 __u32 __user *lmmsizep = NULL;
1762 struct lu_fid __user *fidp = NULL;
1765 if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1766 cmd == IOC_MDC_GETFILEINFO_V2 ||
1767 cmd == IOC_MDC_GETFILESTRIPE) {
1768 filename = ll_getname((const char __user *)arg);
1769 if (IS_ERR(filename))
1770 RETURN(PTR_ERR(filename));
1772 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1773 &lmmsize, &request);
1775 rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1781 body = req_capsule_server_get(&request->rq_pill,
1783 LASSERT(body != NULL);
1788 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO_V1 ||
1789 cmd == LL_IOC_MDC_GETINFO_V1 ||
1790 cmd == IOC_MDC_GETFILEINFO_V2 ||
1791 cmd == LL_IOC_MDC_GETINFO_V2)) {
1799 if (cmd == IOC_MDC_GETFILESTRIPE ||
1800 cmd == LL_IOC_LOV_GETSTRIPE ||
1801 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1802 lump = (struct lov_user_md __user *)arg;
1803 } else if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1804 cmd == LL_IOC_MDC_GETINFO_V1){
1805 struct lov_user_mds_data_v1 __user *lmdp;
1807 lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1808 statp = &lmdp->lmd_st;
1809 lump = &lmdp->lmd_lmm;
1811 struct lov_user_mds_data __user *lmdp;
1813 lmdp = (struct lov_user_mds_data __user *)arg;
1814 fidp = &lmdp->lmd_fid;
1815 stxp = &lmdp->lmd_stx;
1816 flagsp = &lmdp->lmd_flags;
1817 lmmsizep = &lmdp->lmd_lmmsize;
1818 lump = &lmdp->lmd_lmm;
1822 /* If the file has no striping then zero out *lump so
1823 * that the caller isn't confused by garbage. */
1824 if (clear_user(lump, sizeof(*lump)))
1825 GOTO(out_req, rc = -EFAULT);
1826 } else if (copy_to_user(lump, lmm, lmmsize)) {
1827 if (copy_to_user(lump, lmm, sizeof(*lump)))
1828 GOTO(out_req, rc = -EFAULT);
1832 if (cmd == IOC_MDC_GETFILEINFO_V1 ||
1833 cmd == LL_IOC_MDC_GETINFO_V1) {
1836 st.st_dev = inode->i_sb->s_dev;
1837 st.st_mode = body->mbo_mode;
1838 st.st_nlink = body->mbo_nlink;
1839 st.st_uid = body->mbo_uid;
1840 st.st_gid = body->mbo_gid;
1841 st.st_rdev = body->mbo_rdev;
1842 st.st_size = body->mbo_size;
1843 st.st_blksize = PAGE_SIZE;
1844 st.st_blocks = body->mbo_blocks;
1845 st.st_atime = body->mbo_atime;
1846 st.st_mtime = body->mbo_mtime;
1847 st.st_ctime = body->mbo_ctime;
1848 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1852 if (copy_to_user(statp, &st, sizeof(st)))
1853 GOTO(out_req, rc = -EFAULT);
1854 } else if (cmd == IOC_MDC_GETFILEINFO_V2 ||
1855 cmd == LL_IOC_MDC_GETINFO_V2) {
1856 lstatx_t stx = { 0 };
1857 __u64 valid = body->mbo_valid;
1859 stx.stx_blksize = PAGE_SIZE;
1860 stx.stx_nlink = body->mbo_nlink;
1861 stx.stx_uid = body->mbo_uid;
1862 stx.stx_gid = body->mbo_gid;
1863 stx.stx_mode = body->mbo_mode;
1864 stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1867 stx.stx_size = body->mbo_size;
1868 stx.stx_blocks = body->mbo_blocks;
1869 stx.stx_atime.tv_sec = body->mbo_atime;
1870 stx.stx_ctime.tv_sec = body->mbo_ctime;
1871 stx.stx_mtime.tv_sec = body->mbo_mtime;
1872 stx.stx_btime.tv_sec = body->mbo_btime;
1873 stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1874 stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1875 stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1876 stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1877 stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
1880 * For a striped directory, the size and blocks returned
1881 * from MDT is not correct.
1882 * The size and blocks are aggregated by client across
1884 * Thus for a striped directory, do not return the valid
1885 * FLSIZE and FLBLOCKS flags to the caller.
1886 * However, this whould be better decided by the MDS
1887 * instead of the client.
1889 if (cmd == LL_IOC_MDC_GETINFO_V2 &&
1890 ll_i2info(inode)->lli_lsm_md != NULL)
1891 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1893 if (flagsp && copy_to_user(flagsp, &valid,
1895 GOTO(out_req, rc = -EFAULT);
1897 if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1899 GOTO(out_req, rc = -EFAULT);
1901 if (!(valid & OBD_MD_FLSIZE))
1902 stx.stx_mask &= ~STATX_SIZE;
1903 if (!(valid & OBD_MD_FLBLOCKS))
1904 stx.stx_mask &= ~STATX_BLOCKS;
1906 if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1907 GOTO(out_req, rc = -EFAULT);
1909 if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1911 GOTO(out_req, rc = -EFAULT);
1916 ptlrpc_req_finished(request);
1917 ptlrpc_req_finished(root_request);
1919 ll_putname(filename);
1922 case OBD_IOC_QUOTACTL: {
1923 struct if_quotactl *qctl;
1924 int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
1926 OBD_ALLOC(qctl, qctl_len);
1930 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1931 GOTO(out_quotactl, rc = -EFAULT);
1933 if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
1934 char __user *from = (char __user *)arg +
1935 offsetof(typeof(*qctl), qc_poolname);
1936 if (copy_from_user(qctl->qc_poolname, from,
1937 LOV_MAXPOOLNAME + 1))
1938 GOTO(out_quotactl, rc = -EFAULT);
1941 rc = quotactl_ioctl(inode->i_sb, qctl);
1943 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1947 OBD_FREE(qctl, qctl_len);
1950 case OBD_IOC_GETNAME_OLD:
1952 case OBD_IOC_GETDTNAME:
1954 case OBD_IOC_GETMDNAME:
1955 RETURN(ll_get_obd_name(inode, cmd, arg));
1956 case LL_IOC_FLUSHCTX:
1957 RETURN(ll_flush_ctx(inode));
1958 case LL_IOC_GETOBDCOUNT: {
1960 struct obd_export *exp;
1962 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1965 /* get ost count when count is zero, get mdt count otherwise */
1966 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1967 vallen = sizeof(count);
1968 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1969 KEY_TGT_COUNT, &vallen, &count);
1971 CERROR("get target count failed: %d\n", rc);
1975 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1980 case LL_IOC_PATH2FID:
1981 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1982 sizeof(struct lu_fid)))
1985 case LL_IOC_GET_CONNECT_FLAGS: {
1986 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1987 (void __user *)arg));
1989 case OBD_IOC_FID2PATH:
1990 RETURN(ll_fid2path(inode, (void __user *)arg));
1991 case LL_IOC_GETPARENT:
1992 RETURN(ll_getparent(file, (void __user *)arg));
1993 case LL_IOC_FID2MDTIDX: {
1994 struct obd_export *exp = ll_i2mdexp(inode);
1998 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
2002 /* Call mdc_iocontrol */
2003 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
2004 (__u32 __user *)&index);
2010 case LL_IOC_HSM_REQUEST: {
2011 struct hsm_user_request *hur;
2018 /* We don't know the true size yet; copy the fixed-size part */
2019 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
2024 /* Compute the whole struct size */
2025 totalsize = hur_len(hur);
2030 /* Final size will be more than double totalsize */
2031 if (totalsize >= MDS_MAXREQSIZE / 3)
2034 OBD_ALLOC_LARGE(hur, totalsize);
2038 /* Copy the whole struct */
2039 if (copy_from_user(hur, (void __user *)arg, totalsize))
2040 GOTO(out_hur, rc = -EFAULT);
2042 if (hur->hur_request.hr_action == HUA_RELEASE) {
2043 const struct lu_fid *fid;
2047 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
2048 fid = &hur->hur_user_item[i].hui_fid;
2049 f = search_inode_for_lustre(inode->i_sb, fid);
2055 rc = ll_hsm_release(f);
2061 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
2066 OBD_FREE_LARGE(hur, totalsize);
2070 case LL_IOC_HSM_PROGRESS: {
2071 struct hsm_progress_kernel hpk;
2072 struct hsm_progress hp;
2074 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
2077 hpk.hpk_fid = hp.hp_fid;
2078 hpk.hpk_cookie = hp.hp_cookie;
2079 hpk.hpk_extent = hp.hp_extent;
2080 hpk.hpk_flags = hp.hp_flags;
2081 hpk.hpk_errval = hp.hp_errval;
2082 hpk.hpk_data_version = 0;
2084 /* File may not exist in Lustre; all progress
2085 * reported to Lustre root */
2086 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2090 case LL_IOC_HSM_CT_START:
2091 if (!capable(CAP_SYS_ADMIN))
2094 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2095 (struct lustre_kernelcomm __user *)arg);
2098 case LL_IOC_HSM_COPY_START: {
2099 struct hsm_copy *copy;
2102 OBD_ALLOC_PTR(copy);
2105 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2110 rc = ll_ioc_copy_start(inode->i_sb, copy);
2111 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2117 case LL_IOC_HSM_COPY_END: {
2118 struct hsm_copy *copy;
2121 OBD_ALLOC_PTR(copy);
2124 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2129 rc = ll_ioc_copy_end(inode->i_sb, copy);
2130 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2136 case LL_IOC_MIGRATE: {
2137 struct lmv_user_md *lum;
2143 rc = obd_ioctl_getdata(&data, &len, (void __user *)arg);
2147 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2148 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2149 GOTO(migrate_free, rc = -EINVAL);
2151 filename = data->ioc_inlbuf1;
2152 namelen = data->ioc_inllen1;
2154 if (namelen < 1 || namelen != strlen(filename) + 1) {
2155 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2156 GOTO(migrate_free, rc = -EINVAL);
2159 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2160 if (lum->lum_magic != LMV_USER_MAGIC &&
2161 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2163 CERROR("%s: wrong lum magic %x: rc = %d\n",
2164 filename, lum->lum_magic, rc);
2165 GOTO(migrate_free, rc);
2168 rc = ll_migrate(inode, file, lum, filename);
2170 OBD_FREE_LARGE(data, len);
2174 case FS_IOC_FSGETXATTR:
2175 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2176 case FS_IOC_FSSETXATTR:
2177 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2178 case LL_IOC_PROJECT:
2179 RETURN(ll_ioctl_project(file, cmd, arg));
2180 case LL_IOC_PCC_DETACH_BY_FID: {
2181 struct lu_pcc_detach_fid *detach;
2183 struct inode *inode2;
2186 OBD_ALLOC_PTR(detach);
2190 if (copy_from_user(detach,
2191 (const struct lu_pcc_detach_fid __user *)arg,
2193 GOTO(out_detach, rc = -EFAULT);
2195 fid = &detach->pccd_fid;
2196 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2197 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2199 /* Target inode is not in inode cache, and PCC file
2200 * has aleady released, return immdiately.
2202 GOTO(out_detach, rc = 0);
2204 if (!S_ISREG(inode2->i_mode))
2205 GOTO(out_iput, rc = -EINVAL);
2207 if (!inode_owner_or_capable(inode2))
2208 GOTO(out_iput, rc = -EPERM);
2210 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2214 OBD_FREE_PTR(detach);
2217 #ifdef HAVE_LUSTRE_CRYPTO
2218 case LL_IOC_SET_ENCRYPTION_POLICY:
2219 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2221 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
2222 case LL_IOC_GET_ENCRYPTION_POLICY_EX:
2223 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2225 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
2226 case LL_IOC_ADD_ENCRYPTION_KEY:
2227 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2229 return llcrypt_ioctl_add_key(file, (void __user *)arg);
2230 case LL_IOC_REMOVE_ENCRYPTION_KEY:
2231 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2233 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
2234 case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
2235 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2237 return llcrypt_ioctl_remove_key_all_users(file,
2238 (void __user *)arg);
2239 case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
2240 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2242 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
2245 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2246 (void __user *)arg));
2250 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2252 struct inode *inode = file->f_mapping->host;
2253 struct ll_file_data *fd = file->private_data;
2254 struct ll_sb_info *sbi = ll_i2sbi(inode);
2255 int api32 = ll_need_32bit_api(sbi);
2256 loff_t ret = -EINVAL;
2264 offset += file->f_pos;
2270 offset += LL_DIR_END_OFF_32BIT;
2272 offset += LL_DIR_END_OFF;
2279 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2280 (!api32 && offset <= LL_DIR_END_OFF))) {
2281 if (offset != file->f_pos) {
2282 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2283 (!api32 && offset == LL_DIR_END_OFF))
2284 fd->lfd_pos = MDS_DIR_END_OFF;
2285 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2286 fd->lfd_pos = offset << 32;
2288 fd->lfd_pos = offset;
2289 file->f_pos = offset;
2290 file->f_version = 0;
2297 inode_unlock(inode);
2301 static int ll_dir_open(struct inode *inode, struct file *file)
2304 RETURN(ll_file_open(inode, file));
2307 static int ll_dir_release(struct inode *inode, struct file *file)
2310 RETURN(ll_file_release(inode, file));
2313 const struct file_operations ll_dir_operations = {
2314 .llseek = ll_dir_seek,
2315 .open = ll_dir_open,
2316 .release = ll_dir_release,
2317 .read = generic_read_dir,
2318 #ifdef HAVE_DIR_CONTEXT
2319 .iterate_shared = ll_iterate,
2321 .readdir = ll_readdir,
2323 .unlocked_ioctl = ll_dir_ioctl,