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) 2003, 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/
32 #ifndef LLITE_INTERNAL_H
33 #define LLITE_INTERNAL_H
35 #include <lustre_disk.h> /* for s2sbi */
36 #include <lustre_linkea.h>
38 /* for struct cl_lock_descr and struct cl_io */
39 #include <cl_object.h>
40 #include <lustre_lmv.h>
41 #include <lustre_mdc.h>
42 #include <lustre_intent.h>
43 #include <linux/compat.h>
44 #include <linux/aio.h>
45 #include <linux/parser.h>
46 #include <linux/seqlock.h>
47 #include <lustre_compat.h>
48 #include <lustre_crypto.h>
49 #include <range_lock.h>
51 #include "vvp_internal.h"
53 #include "foreign_symlink.h"
59 #ifndef HAVE_VM_FAULT_RETRY
60 #define VM_FAULT_RETRY 0
63 /* Kernel 3.1 kills LOOKUP_CONTINUE, LOOKUP_PARENT is equivalent to it.
64 * seem kernel commit 49084c3bb2055c401f3493c13edae14d49128ca0 */
65 #ifndef LOOKUP_CONTINUE
66 #define LOOKUP_CONTINUE LOOKUP_PARENT
69 #ifdef HAVE_DENTRY_D_U_D_ALIAS
70 #define d_alias d_u.d_alias
73 /** Only used on client-side for indicating the tail of dir hash/offset. */
74 #define LL_DIR_END_OFF 0x7fffffffffffffffULL
75 #define LL_DIR_END_OFF_32BIT 0x7fffffffUL
77 /* 4UL * 1024 * 1024 */
78 #define LL_MAX_BLKSIZE_BITS 22
80 #define LL_IT2STR(it) ((it) ? ldlm_it2str((it)->it_op) : "0")
82 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
84 struct ll_dentry_data {
85 unsigned int lld_sa_generation;
86 unsigned int lld_invalid:1;
87 unsigned int lld_nfs_dentry:1;
88 struct rcu_head lld_rcu_head;
91 #define ll_d2d(de) ((struct ll_dentry_data*)((de)->d_fsdata))
93 #define LLI_INODE_MAGIC 0x111d0de5
94 #define LLI_INODE_DEAD 0xdeadd00d
96 struct ll_getname_data {
97 #ifdef HAVE_DIR_CONTEXT
98 struct dir_context ctx;
100 char *lgd_name; /* points to a buffer with NAME_MAX+1 size */
101 struct lu_fid lgd_fid; /* target fid we are looking for */
102 int lgd_found; /* inode matched? */
105 struct ll_grouplock {
106 struct lu_env *lg_env;
108 struct cl_lock *lg_lock;
109 unsigned long lg_gid;
112 /* See comment on trunc_sem_down_read_nowait */
113 struct ll_trunc_sem {
114 /* when positive, this is a count of readers, when -1, it indicates
115 * the semaphore is held for write, and 0 is unlocked
117 atomic_t ll_trunc_readers;
118 /* this tracks a count of waiting writers */
119 atomic_t ll_trunc_waiters;
122 struct ll_inode_info {
123 __u32 lli_inode_magic;
126 volatile unsigned long lli_flags;
127 struct posix_acl *lli_posix_acl;
129 /* identifying fields for both metadata and data stacks. */
130 struct lu_fid lli_fid;
131 /* master inode fid for stripe directory */
132 struct lu_fid lli_pfid;
134 /* We need all three because every inode may be opened in different
136 struct obd_client_handle *lli_mds_read_och;
137 struct obd_client_handle *lli_mds_write_och;
138 struct obd_client_handle *lli_mds_exec_och;
139 __u64 lli_open_fd_read_count;
140 __u64 lli_open_fd_write_count;
141 __u64 lli_open_fd_exec_count;
143 /* Number of times this inode was opened */
144 u64 lli_open_fd_count;
145 /* When last close was performed on this inode */
146 ktime_t lli_close_fd_time;
148 /* Protects access to och pointers and their usage counters */
149 struct mutex lli_och_mutex;
151 struct inode lli_vfs_inode;
153 /* the most recent timestamps obtained from mds */
158 spinlock_t lli_agl_lock;
160 /* Try to make the d::member and f::member are aligned. Before using
161 * these members, make clear whether it is directory or not. */
165 /* metadata statahead */
166 /* since parent-child threads can share the same @file
167 * struct, "opendir_key" is the token when dir close for
168 * case of parent exit before child -- it is me should
169 * cleanup the dir readahead. */
170 void *lli_opendir_key;
171 struct ll_statahead_info *lli_sai;
172 /* protect statahead stuff. */
173 spinlock_t lli_sa_lock;
174 /* "opendir_pid" is the token when lookup/revalid
175 * -- I am the owner of dir statahead. */
176 pid_t lli_opendir_pid;
177 /* directory depth to ROOT */
178 unsigned short lli_dir_depth;
179 /* directory depth to ancestor whose default LMV is
182 unsigned short lli_inherit_depth;
183 /* stat will try to access statahead entries or start
184 * statahead if this flag is set, and this flag will be
185 * set upon dir open, and cleared when dir is closed,
186 * statahead hit ratio is too low, or start statahead
188 unsigned short lli_sa_enabled:1;
189 /* generation for statahead */
190 unsigned int lli_sa_generation;
191 /* rw lock protects lli_lsm_md */
192 struct rw_semaphore lli_lsm_sem;
193 /* directory stripe information */
194 struct lmv_stripe_md *lli_lsm_md;
195 /* directory default LMV */
196 struct lmv_stripe_md *lli_default_lsm_md;
199 /* for non-directory */
201 struct mutex lli_size_mutex;
202 char *lli_symlink_name;
203 struct ll_trunc_sem lli_trunc_sem;
204 struct range_lock_tree lli_write_tree;
205 struct mutex lli_setattr_mutex;
207 struct rw_semaphore lli_glimpse_sem;
208 ktime_t lli_glimpse_time;
209 struct list_head lli_agl_list;
212 /* for writepage() only to communicate to fsync */
215 /* protect the file heat fields */
216 spinlock_t lli_heat_lock;
217 __u32 lli_heat_flags;
218 struct obd_heat_instance lli_heat_instances[OBD_HEAT_COUNT];
221 * Whenever a process try to read/write the file, the
222 * jobid of the process will be saved here, and it'll
223 * be packed into the write PRC when flush later.
225 * So the read/write statistics for jobid will not be
226 * accurate if the file is shared by different jobs.
228 char lli_jobid[LUSTRE_JOBID_SIZE];
230 struct mutex lli_pcc_lock;
231 enum lu_pcc_state_flags lli_pcc_state;
233 * @lli_pcc_generation saves the gobal PCC generation
234 * when the file was successfully attached into PCC.
235 * The flags of the PCC dataset are saved in
237 * The gobal PCC generation will be increased when add
238 * or delete a PCC backend, or change the configuration
239 * parameters for PCC.
240 * If @lli_pcc_generation is same as the gobal PCC
241 * generation, we can use the saved flags of the PCC
242 * dataset to determine whether need to try auto attach
245 __u64 lli_pcc_generation;
246 enum pcc_dataset_flags lli_pcc_dsflags;
247 struct pcc_inode *lli_pcc_inode;
249 struct mutex lli_group_mutex;
250 __u64 lli_group_users;
251 unsigned long lli_group_gid;
253 __u64 lli_attr_valid;
255 __u64 lli_lazyblocks;
259 /* XXX: For following frequent used members, although they maybe special
260 * used for non-directory object, it is some time-wasting to check
261 * whether the object is directory or not before using them. On the
262 * other hand, currently, sizeof(f) > sizeof(d), it cannot reduce
263 * the "ll_inode_info" size even if moving those members into u.f.
264 * So keep them out side.
266 * In the future, if more members are added only for directory,
267 * some of the following members can be moved into u.f.
269 struct cl_object *lli_clob;
271 /* mutex to request for layout lock exclusively. */
272 struct mutex lli_layout_mutex;
273 /* Layout version, protected by lli_layout_lock */
274 __u32 lli_layout_gen;
275 spinlock_t lli_layout_lock;
277 __u32 lli_projid; /* project id */
279 struct rw_semaphore lli_xattrs_list_rwsem;
280 struct mutex lli_xattrs_enq_lock;
281 struct list_head lli_xattrs; /* ll_xattr_entry->xe_list */
282 struct list_head lli_lccs; /* list of ll_cl_context */
283 seqlock_t lli_page_inv_lock;
286 #ifndef HAVE_USER_NAMESPACE_ARG
287 #define inode_permission(ns, inode, mask) inode_permission(inode, mask)
288 #define generic_permission(ns, inode, mask) generic_permission(inode, mask)
289 #define simple_setattr(ns, de, iattr) simple_setattr(de, iattr)
290 #define ll_inode_permission(ns, inode, mask) ll_inode_permission(inode, mask)
291 #ifdef HAVE_INODEOPS_ENHANCED_GETATTR
292 #define ll_getattr(ns, path, stat, mask, fl) ll_getattr(path, stat, mask, fl)
293 #endif /* HAVE_INODEOPS_ENHANCED_GETATTR */
294 #define ll_setattr(ns, de, attr) ll_setattr(de, attr)
297 static inline void ll_trunc_sem_init(struct ll_trunc_sem *sem)
299 atomic_set(&sem->ll_trunc_readers, 0);
300 atomic_set(&sem->ll_trunc_waiters, 0);
303 /* This version of down read ignores waiting writers, meaning if the semaphore
304 * is already held for read, this down_read will 'join' that reader and also
305 * take the semaphore.
307 * This lets us avoid an unusual deadlock.
309 * We must take lli_trunc_sem in read mode on entry in to various i/o paths
310 * in Lustre, in order to exclude truncates. Some of these paths then need to
311 * take the mmap_lock, while still holding the trunc_sem. The problem is that
312 * page faults hold the mmap_lock when calling in to Lustre, and then must also
313 * take the trunc_sem to exclude truncate.
315 * This means the locking order for trunc_sem and mmap_lock is sometimes AB,
316 * sometimes BA. This is almost OK because in both cases, we take the trunc
317 * sem for read, so it doesn't block.
319 * However, if a write mode user (truncate, a setattr op) arrives in the
320 * middle of this, the second reader on the truncate_sem will wait behind that
323 * So we have, on our truncate sem, in order (where 'reader' and 'writer' refer
324 * to the mode in which they take the semaphore):
325 * reader (holding mmap_lock, needs truncate_sem)
327 * reader (holding truncate sem, waiting for mmap_lock)
329 * And so the readers deadlock.
331 * The solution is this modified semaphore, where this down_read ignores
332 * waiting write operations, and all waiters are woken up at once, so readers
333 * using down_read_nowait cannot get stuck behind waiting writers, regardless
334 * of the order they arrived in.
336 * down_read_nowait is only used in the page fault case, where we already hold
337 * the mmap_lock. This is because otherwise repeated read and write operations
338 * (which take the truncate sem) could prevent a truncate from ever starting.
339 * This could still happen with page faults, but without an even more complex
340 * mechanism, this is unavoidable.
344 static inline void trunc_sem_down_read_nowait(struct ll_trunc_sem *sem)
346 wait_var_event(&sem->ll_trunc_readers,
347 atomic_inc_unless_negative(&sem->ll_trunc_readers));
350 static inline void trunc_sem_down_read(struct ll_trunc_sem *sem)
352 wait_var_event(&sem->ll_trunc_readers,
353 atomic_read(&sem->ll_trunc_waiters) == 0 &&
354 atomic_inc_unless_negative(&sem->ll_trunc_readers));
357 static inline void trunc_sem_up_read(struct ll_trunc_sem *sem)
359 if (atomic_dec_return(&sem->ll_trunc_readers) == 0 &&
360 atomic_read(&sem->ll_trunc_waiters))
361 wake_up_var(&sem->ll_trunc_readers);
364 static inline void trunc_sem_down_write(struct ll_trunc_sem *sem)
366 atomic_inc(&sem->ll_trunc_waiters);
367 wait_var_event(&sem->ll_trunc_readers,
368 atomic_cmpxchg(&sem->ll_trunc_readers, 0, -1) == 0);
369 atomic_dec(&sem->ll_trunc_waiters);
372 static inline void trunc_sem_up_write(struct ll_trunc_sem *sem)
374 atomic_set(&sem->ll_trunc_readers, 0);
375 /* match the smp_mb() in wait_var_event()->prepare_to_wait() */
377 wake_up_var(&sem->ll_trunc_readers);
380 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
381 static inline void lli_clear_acl(struct ll_inode_info *lli)
383 if (lli->lli_posix_acl) {
384 posix_acl_release(lli->lli_posix_acl);
385 lli->lli_posix_acl = NULL;
389 static inline void lli_replace_acl(struct ll_inode_info *lli,
390 struct lustre_md *md)
392 write_lock(&lli->lli_lock);
393 if (lli->lli_posix_acl)
394 posix_acl_release(lli->lli_posix_acl);
395 lli->lli_posix_acl = md->posix_acl;
396 write_unlock(&lli->lli_lock);
399 static inline void lli_clear_acl(struct ll_inode_info *lli)
403 static inline void lli_replace_acl(struct ll_inode_info *lli,
404 struct lustre_md *md)
409 static inline __u32 ll_layout_version_get(struct ll_inode_info *lli)
413 spin_lock(&lli->lli_layout_lock);
414 gen = lli->lli_layout_gen;
415 spin_unlock(&lli->lli_layout_lock);
420 static inline void ll_layout_version_set(struct ll_inode_info *lli, __u32 gen)
422 spin_lock(&lli->lli_layout_lock);
423 lli->lli_layout_gen = gen;
424 spin_unlock(&lli->lli_layout_lock);
428 /* File data is modified. */
429 LLIF_DATA_MODIFIED = 0,
430 /* File is being restored */
431 LLIF_FILE_RESTORING = 1,
432 /* Xattr cache is attached to the file */
433 LLIF_XATTR_CACHE = 2,
434 /* Project inherit */
435 LLIF_PROJECT_INHERIT = 3,
436 /* update atime from MDS even if it's older than local inode atime. */
437 LLIF_UPDATE_ATIME = 4,
438 /* foreign file/dir can be unlinked unconditionnaly */
439 LLIF_FOREIGN_REMOVABLE = 5,
440 /* Xattr cache is filled */
441 LLIF_XATTR_CACHE_FILLED = 7,
445 int ll_xattr_cache_destroy(struct inode *inode);
446 int ll_xattr_cache_empty(struct inode *inode);
448 int ll_xattr_cache_get(struct inode *inode,
454 int ll_xattr_cache_insert(struct inode *inode,
459 static inline bool obd_connect_has_secctx(struct obd_connect_data *data)
461 #ifdef CONFIG_SECURITY
462 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
463 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
469 static inline void obd_connect_set_secctx(struct obd_connect_data *data)
471 #ifdef CONFIG_SECURITY
472 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
476 /* Only smack and selinux is known to use security contexts */
477 static inline bool ll_xattr_is_seclabel(const char *name)
479 return !strcmp(name, XATTR_NAME_SELINUX) ||
480 !strcmp(name, XATTR_NAME_SMACK);
483 static inline bool ll_xattr_suffix_is_seclabel(const char *suffix)
485 return !strcmp(suffix, XATTR_SELINUX_SUFFIX) ||
486 !strcmp(suffix, XATTR_SMACK_SUFFIX);
489 int ll_dentry_init_security(struct dentry *dentry, int mode, struct qstr *name,
490 const char **secctx_name, __u32 *secctx_name_size,
491 void **secctx, __u32 *secctx_size);
492 int ll_inode_init_security(struct dentry *dentry, struct inode *inode,
495 int ll_inode_notifysecctx(struct inode *inode,
496 void *secctx, __u32 secctxlen);
498 void ll_secctx_name_free(struct ll_sb_info *sbi);
500 int ll_secctx_name_store(struct inode *in);
502 __u32 ll_secctx_name_get(struct ll_sb_info *sbi, const char **secctx_name);
504 int ll_security_secctx_name_filter(struct ll_sb_info *sbi, int xattr_type,
507 static inline bool obd_connect_has_enc(struct obd_connect_data *data)
509 #ifdef HAVE_LUSTRE_CRYPTO
510 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
511 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT;
517 static inline void obd_connect_set_enc(struct obd_connect_data *data)
519 #ifdef HAVE_LUSTRE_CRYPTO
520 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT;
524 static inline bool obd_connect_has_name_enc(struct obd_connect_data *data)
526 #ifdef HAVE_LUSTRE_CRYPTO
527 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
528 data->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT_NAME;
534 static inline void obd_connect_set_name_enc(struct obd_connect_data *data)
536 #ifdef HAVE_LUSTRE_CRYPTO
537 data->ocd_connect_flags2 |= OBD_CONNECT2_ENCRYPT_NAME;
542 * Locking to guarantee consistency of non-atomic updates to long long i_size,
543 * consistency between file size and KMS.
545 * Implemented by ->lli_size_mutex and ->lsm_lock, nested in that order.
548 void ll_inode_size_lock(struct inode *inode);
549 void ll_inode_size_unlock(struct inode *inode);
551 static inline struct ll_inode_info *ll_i2info(struct inode *inode)
553 return container_of(inode, struct ll_inode_info, lli_vfs_inode);
556 static inline struct pcc_inode *ll_i2pcci(struct inode *inode)
558 return ll_i2info(inode)->lli_pcc_inode;
561 /* default to use at least 16M for fast read if possible */
562 #define RA_REMAIN_WINDOW_MIN MiB_TO_PAGES(16UL)
564 /* default read-ahead on a given client mountpoint. */
565 #define SBI_DEFAULT_READ_AHEAD_MAX MiB_TO_PAGES(1024UL)
567 /* default read-ahead for a single file descriptor */
568 #define SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX MiB_TO_PAGES(256UL)
570 /* default read-ahead full files smaller than limit on the second read */
571 #define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX MiB_TO_PAGES(2UL)
573 /* default range pages */
574 #define SBI_DEFAULT_RA_RANGE_PAGES MiB_TO_PAGES(1ULL)
576 /* Min range pages */
577 #define RA_MIN_MMAP_RANGE_PAGES 16UL
582 RA_STAT_DISTANT_READPAGE,
583 RA_STAT_MISS_IN_WINDOW,
584 RA_STAT_FAILED_GRAB_PAGE,
585 RA_STAT_FAILED_MATCH,
590 RA_STAT_MAX_IN_FLIGHT,
591 RA_STAT_WRONG_GRAB_PAGE,
592 RA_STAT_FAILED_REACH_END,
594 RA_STAT_FAILED_FAST_READ,
595 RA_STAT_MMAP_RANGE_READ,
596 RA_STAT_READAHEAD_PAGES,
601 atomic_t ra_cur_pages;
602 unsigned long ra_max_pages;
603 unsigned long ra_max_pages_per_file;
604 unsigned long ra_range_pages;
605 unsigned long ra_max_read_ahead_whole_pages;
606 struct workqueue_struct *ll_readahead_wq;
608 * Max number of active works could be triggered
609 * for async readahead.
611 unsigned int ra_async_max_active;
612 /* how many async readahead triggered in flight */
613 atomic_t ra_async_inflight;
614 /* Threshold to control when to trigger async readahead */
615 unsigned long ra_async_pages_per_file_threshold;
618 /* ra_io_arg will be filled in the beginning of ll_readahead with
619 * ras_lock, then the following ll_read_ahead_pages will read RA
620 * pages according to this arg, all the items in this structure are
621 * counted by page index.
624 pgoff_t ria_start_idx; /* start offset of read-ahead*/
625 pgoff_t ria_end_idx; /* end offset of read-ahead*/
626 unsigned long ria_reserved; /* reserved pages for read-ahead */
627 pgoff_t ria_end_idx_min;/* minimum end to cover current read */
628 bool ria_eof; /* reach end of file */
629 /* If stride read pattern is detected, ria_stoff is the byte offset
630 * where stride read is started. Note: for normal read-ahead, the
631 * value here is meaningless, and also it will not be accessed*/
633 /* ria_length and ria_bytes are the length and pages length in the
634 * stride I/O mode. And they will also be used to check whether
635 * it is stride I/O read-ahead in the read-ahead pages*/
640 /* LL_HIST_MAX=32 causes an overflow */
641 #define LL_HIST_MAX 28
642 #define LL_HIST_START 12 /* buckets start at 2^12 = 4k */
643 #define LL_PROCESS_HIST_MAX 10
644 struct per_process_info {
646 struct obd_histogram pp_r_hist;
647 struct obd_histogram pp_w_hist;
650 /* pp_extents[LL_PROCESS_HIST_MAX] will hold the combined process info */
651 struct ll_rw_extents_info {
653 struct per_process_info pp_extents[LL_PROCESS_HIST_MAX + 1];
656 #define LL_OFFSET_HIST_MAX 100
657 struct ll_rw_process_info {
660 loff_t rw_range_start;
662 loff_t rw_last_file_pos;
664 size_t rw_smallest_extent;
665 size_t rw_largest_extent;
666 struct ll_file_data *rw_last_file;
669 enum stats_track_type {
670 STATS_TRACK_ALL = 0, /* track all processes */
671 STATS_TRACK_PID, /* track process with this pid */
672 STATS_TRACK_PPID, /* track processes with this ppid */
673 STATS_TRACK_GID, /* track processes with this gid */
677 /* flags for sbi->ll_flags */
679 LL_SBI_NOLCK, /* DLM locking disabled directio-only */
680 LL_SBI_CHECKSUM, /* checksum each page as it's written */
681 LL_SBI_LOCALFLOCK, /* local flocks instead of fs-wide */
682 LL_SBI_FLOCK, /* flock enabled */
683 LL_SBI_USER_XATTR, /* support user xattr */
684 LL_SBI_LRU_RESIZE, /* lru resize support */
685 LL_SBI_LAZYSTATFS, /* lazystatfs mount option */
686 LL_SBI_32BIT_API, /* generate 32 bit inodes. */
687 LL_SBI_USER_FID2PATH, /* fid2path by unprivileged users */
688 LL_SBI_VERBOSE, /* verbose mount/umount */
689 LL_SBI_ALWAYS_PING, /* ping even if server suppress_pings */
690 LL_SBI_TEST_DUMMY_ENCRYPTION, /* test dummy encryption */
691 LL_SBI_ENCRYPT, /* client side encryption */
692 LL_SBI_FOREIGN_SYMLINK, /* foreign fake-symlink support */
693 LL_SBI_FOREIGN_SYMLINK_UPCALL, /* foreign fake-symlink upcall set */
694 LL_SBI_NUM_MOUNT_OPT,
696 LL_SBI_ACL, /* support ACL */
697 LL_SBI_AGL_ENABLED, /* enable agl */
698 LL_SBI_64BIT_HASH, /* support 64-bits dir hash/offset */
699 LL_SBI_LAYOUT_LOCK, /* layout lock support */
700 LL_SBI_XATTR_CACHE, /* support for xattr cache */
701 LL_SBI_NOROOTSQUASH, /* do not apply root squash */
702 LL_SBI_FAST_READ, /* fast read support */
703 LL_SBI_FILE_SECCTX, /* file security context at create */
704 LL_SBI_TINY_WRITE, /* tiny write support */
705 LL_SBI_FILE_HEAT, /* file heat support */
706 LL_SBI_PARALLEL_DIO, /* parallel (async) O_DIRECT RPCs */
707 LL_SBI_ENCRYPT_NAME, /* name encryption */
711 int ll_sbi_flags_seq_show(struct seq_file *m, void *v);
713 /* This is embedded into llite super-blocks to keep track of connect
714 * flags (capabilities) supported by all imports given mount is
716 struct lustre_client_ocd {
717 /* This is conjunction of connect_flags across all imports
718 * (LOVs) this mount is connected to. This field is updated by
719 * cl_ocd_update() under ->lco_lock. */
721 struct mutex lco_lock;
722 struct obd_export *lco_md_exp;
723 struct obd_export *lco_dt_exp;
727 /* this protects pglist and ra_info. It isn't safe to
728 * grab from interrupt contexts */
730 spinlock_t ll_pp_extent_lock; /* pp_extent entry*/
731 spinlock_t ll_process_lock; /* ll_rw_process_info */
732 struct obd_uuid ll_sb_uuid;
733 struct obd_export *ll_md_exp;
734 struct obd_export *ll_dt_exp;
735 struct obd_device *ll_md_obd;
736 struct obd_device *ll_dt_obd;
737 struct dentry *ll_debugfs_entry;
738 struct lu_fid ll_root_fid; /* root object fid */
739 struct mnt_namespace *ll_mnt_ns;
741 DECLARE_BITMAP(ll_flags, LL_SBI_NUM_FLAGS); /* enum ll_sbi_flags */
742 unsigned int ll_xattr_cache_enabled:1,
743 ll_xattr_cache_set:1, /* already set to 0/1 */
744 ll_client_common_fill_super_succeeded:1,
746 ll_inode_cache_enabled:1;
748 struct lustre_client_ocd ll_lco;
750 struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
752 /* Used to track "unstable" pages on a client, and maintain a
753 * LRU list of clean pages. An "unstable" page is defined as
754 * any page which is sent to a server as part of a bulk request,
755 * but is uncommitted to stable storage. */
756 struct cl_client_cache *ll_cache;
758 struct lprocfs_stats *ll_ra_stats;
760 struct ll_ra_info ll_ra_info;
761 unsigned int ll_namelen;
762 const struct file_operations *ll_fop;
764 struct lu_site *ll_site;
765 struct cl_device *ll_cl;
768 struct ll_rw_extents_info *ll_rw_extents_info;
769 int ll_extent_process_count;
770 unsigned int ll_offset_process_count;
771 struct ll_rw_process_info *ll_rw_process_info;
772 struct ll_rw_process_info *ll_rw_offset_info;
773 ktime_t ll_process_stats_init;
774 unsigned int ll_rw_offset_entry_count;
775 int ll_stats_track_id;
776 enum stats_track_type ll_stats_track_type;
779 /* metadata stat-ahead */
780 unsigned int ll_sa_running_max;/* max concurrent
781 * statahead instances */
782 unsigned int ll_sa_max; /* max statahead RPCs */
783 atomic_t ll_sa_total; /* statahead thread started
785 atomic_t ll_sa_wrong; /* statahead thread stopped for
787 atomic_t ll_sa_running; /* running statahead thread
789 atomic_t ll_agl_total; /* AGL thread started count */
790 atomic_t ll_sa_hit_total; /* total hit count */
791 atomic_t ll_sa_miss_total; /* total miss count */
793 dev_t ll_sdev_orig; /* save s_dev before assign for
796 struct root_squash_info ll_squash;
799 /* st_blksize returned by stat(2), when non-zero */
800 unsigned int ll_stat_blksize;
802 /* maximum relative age of cached statfs results */
803 unsigned int ll_statfs_max_age;
805 struct kset ll_kset; /* sysfs object */
806 struct completion ll_kobj_unregister;
809 unsigned int ll_heat_decay_weight;
810 unsigned int ll_heat_period_second;
812 /* Opens of the same inode before we start requesting open lock */
813 u32 ll_oc_thrsh_count;
815 /* Time in ms between last inode close and next open to be considered
816 * instant back to back and would trigger an open lock request
820 /* Time in ms after last file close that we no longer count prior opens*/
823 /* filesystem fsname */
824 char ll_fsname[LUSTRE_MAXFSNAME + 1];
826 /* Persistent Client Cache */
827 struct pcc_super ll_pcc_super;
829 /* to protect vs updates in all following foreign symlink fields */
830 struct rw_semaphore ll_foreign_symlink_sem;
831 /* foreign symlink path prefix */
832 char *ll_foreign_symlink_prefix;
833 /* full prefix size including leading '\0' */
834 size_t ll_foreign_symlink_prefix_size;
835 /* foreign symlink path upcall */
836 char *ll_foreign_symlink_upcall;
837 /* foreign symlink path upcall infos */
838 struct ll_foreign_symlink_upcall_item *ll_foreign_symlink_upcall_items;
839 /* foreign symlink path upcall nb infos */
840 unsigned int ll_foreign_symlink_upcall_nb_items;
842 /* cached file security context xattr name. e.g: security.selinux */
843 char *ll_secctx_name;
844 __u32 ll_secctx_name_size;
847 #define SBI_DEFAULT_HEAT_DECAY_WEIGHT ((80 * 256 + 50) / 100)
848 #define SBI_DEFAULT_HEAT_PERIOD_SECOND (60)
850 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT (5)
851 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MS (100) /* 0.1 second */
852 #define SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS (60000) /* 1 minute */
855 * per file-descriptor read-ahead data.
857 struct ll_readahead_state {
859 /* End byte that read(2) try to read. */
860 loff_t ras_last_read_end_bytes;
862 * number of bytes read after last read-ahead window reset. As window
863 * is reset on each seek, this is effectively a number of consecutive
864 * accesses. Maybe ->ras_accessed_in_window is better name.
866 * XXX nikita: window is also reset (by ras_update()) when Lustre
867 * believes that memory pressure evicts read-ahead pages. In that
868 * case, it probably doesn't make sense to expand window to
869 * PTLRPC_MAX_BRW_PAGES on the third access.
871 loff_t ras_consecutive_bytes;
873 * number of read requests after the last read-ahead window reset
874 * As window is reset on each seek, this is effectively the number
875 * on consecutive read request and is used to trigger read-ahead.
877 unsigned long ras_consecutive_requests;
879 * Parameters of current read-ahead window. Handled by
880 * ras_update(). On the initial access to the file or after a seek,
881 * window is reset to 0. After 3 consecutive accesses, window is
882 * expanded to PTLRPC_MAX_BRW_PAGES. Afterwards, window is enlarged by
883 * PTLRPC_MAX_BRW_PAGES chunks up to ->ra_max_pages.
885 pgoff_t ras_window_start_idx;
886 pgoff_t ras_window_pages;
888 /* Page index where min range read starts */
889 pgoff_t ras_range_min_start_idx;
890 /* Page index where mmap range read ends */
891 pgoff_t ras_range_max_end_idx;
892 /* number of mmap pages where last time detected */
893 pgoff_t ras_last_range_pages;
894 /* number of mmap range requests */
895 pgoff_t ras_range_requests;
898 * Optimal RPC size in pages.
899 * It decides how many pages will be sent for each read-ahead.
901 unsigned long ras_rpc_pages;
903 * Where next read-ahead should start at. This lies within read-ahead
904 * window. Read-ahead window is read in pieces rather than at once
905 * because: 1. lustre limits total number of pages under read-ahead by
906 * ->ra_max_pages (see ll_ra_count_get()), 2. client cannot read pages
907 * not covered by DLM lock.
909 pgoff_t ras_next_readahead_idx;
911 * Total number of ll_file_read requests issued, reads originating
912 * due to mmap are not counted in this total. This value is used to
913 * trigger full file read-ahead after multiple reads to a small file.
915 unsigned long ras_requests;
917 * The following 3 items are used for detecting the stride I/O
919 * In stride I/O mode,
920 * ...............|-----data-----|****gap*****|--------|******|....
921 * offset |-stride_bytes-|-stride_gap-|
922 * ras_stride_offset = offset;
923 * ras_stride_length = stride_bytes + stride_gap;
924 * ras_stride_bytes = stride_bytes;
925 * Note: all these three items are counted by bytes.
927 loff_t ras_stride_offset;
928 loff_t ras_stride_length;
929 loff_t ras_stride_bytes;
931 * number of consecutive stride request count, and it is similar as
932 * ras_consecutive_requests, but used for stride I/O mode.
933 * Note: only more than 2 consecutive stride request are detected,
934 * stride read-ahead will be enable
936 unsigned long ras_consecutive_stride_requests;
937 /* index of the last page that async readahead starts */
938 pgoff_t ras_async_last_readpage_idx;
939 /* whether we should increase readahead window */
940 bool ras_need_increase_window;
941 /* whether ra miss check should be skipped */
942 bool ras_no_miss_check;
945 struct ll_readahead_work {
946 /** File to readahead */
947 struct file *lrw_file;
948 pgoff_t lrw_start_idx;
952 /* async worker to handler read */
953 struct work_struct lrw_readahead_work;
954 char lrw_jobid[LUSTRE_JOBID_SIZE];
957 extern struct kmem_cache *ll_file_data_slab;
958 struct lustre_handle;
959 struct ll_file_data {
960 struct ll_readahead_state fd_ras;
961 struct ll_grouplock fd_grouplock;
965 /* openhandle if lease exists for this file.
966 * Borrow lli->lli_och_mutex to protect assignment */
967 struct obd_client_handle *fd_lease_och;
968 struct obd_client_handle *fd_och;
969 struct file *fd_file;
970 /* Indicate whether need to report failure when close.
971 * true: failure is known, not report again.
972 * false: unknown failure, should report. */
973 bool fd_write_failed;
974 bool ll_lock_no_expand;
975 /* Used by mirrored file to lead IOs to a specific mirror, usually
976 * for mirror resync. 0 means default. */
977 __u32 fd_designated_mirror;
978 /* The layout version when resync starts. Resync I/O should carry this
979 * layout version for verification to OST objects */
980 __u32 fd_layout_version;
981 struct pcc_file fd_pcc_file;
982 /* striped directory may read partially if some stripe inaccessible,
983 * -errno is saved here, and will return to user in close().
985 int fd_partial_readdir_rc;
988 void llite_tunables_unregister(void);
989 int llite_tunables_register(void);
991 static inline struct inode *ll_info2i(struct ll_inode_info *lli)
993 return &lli->lli_vfs_inode;
996 __u32 ll_i2suppgid(struct inode *i);
997 void ll_i2gids(__u32 *suppgids, struct inode *i1,struct inode *i2);
999 static inline int ll_need_32bit_api(struct ll_sb_info *sbi)
1001 #if BITS_PER_LONG == 32
1003 #elif defined(CONFIG_COMPAT)
1004 if (unlikely(test_bit(LL_SBI_32BIT_API, sbi->ll_flags)))
1007 # ifdef CONFIG_X86_X32
1008 /* in_compat_syscall() returns true when called from a kthread
1009 * and CONFIG_X86_X32 is enabled, which is wrong. So check
1010 * whether the caller comes from a syscall (ie. not a kthread)
1011 * before calling in_compat_syscall(). */
1012 if (current->flags & PF_KTHREAD)
1016 return unlikely(in_compat_syscall());
1018 return unlikely(test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1022 static inline bool ll_sbi_has_fast_read(struct ll_sb_info *sbi)
1024 return test_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1027 static inline bool ll_sbi_has_tiny_write(struct ll_sb_info *sbi)
1029 return test_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1032 static inline bool ll_sbi_has_file_heat(struct ll_sb_info *sbi)
1034 return test_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1037 static inline bool ll_sbi_has_foreign_symlink(struct ll_sb_info *sbi)
1039 return test_bit(LL_SBI_FOREIGN_SYMLINK, sbi->ll_flags);
1042 static inline bool ll_sbi_has_parallel_dio(struct ll_sb_info *sbi)
1044 return test_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1047 void ll_ras_enter(struct file *f, loff_t pos, size_t count);
1049 /* llite/lcommon_misc.c */
1050 int cl_ocd_update(struct obd_device *host, struct obd_device *watched,
1051 enum obd_notify_event ev, void *owner);
1052 int cl_get_grouplock(struct cl_object *obj, unsigned long gid, int nonblock,
1053 struct ll_grouplock *lg);
1054 void cl_put_grouplock(struct ll_grouplock *lg);
1056 /* llite/lproc_llite.c */
1057 int ll_debugfs_register_super(struct super_block *sb, const char *name);
1058 void ll_debugfs_unregister_super(struct super_block *sb);
1059 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count);
1060 void ll_free_rw_stats_info(struct ll_sb_info *sbi);
1063 LPROC_LL_READ_BYTES,
1064 LPROC_LL_WRITE_BYTES,
1091 LPROC_LL_GETXATTR_HITS,
1093 LPROC_LL_REMOVEXATTR,
1094 LPROC_LL_INODE_PERM,
1096 LPROC_LL_INODE_OCOUNT,
1097 LPROC_LL_INODE_OPCLTM,
1098 LPROC_LL_FILE_OPCODES
1102 enum get_default_layout_type {
1103 GET_DEFAULT_LAYOUT_ROOT = 1,
1106 extern const struct file_operations ll_dir_operations;
1107 extern const struct inode_operations ll_dir_inode_operations;
1108 #ifdef HAVE_DIR_CONTEXT
1109 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1110 struct dir_context *ctx, int *partial_readdir_rc);
1112 int ll_dir_read(struct inode *inode, __u64 *pos, struct md_op_data *op_data,
1113 void *cookie, filldir_t filldir, int *partial_readdir_rc);
1115 int ll_get_mdt_idx(struct inode *inode);
1116 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid);
1117 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
1118 __u64 offset, int *partial_readdir_rc);
1119 void ll_release_page(struct inode *inode, struct page *page, bool remove);
1120 int quotactl_ioctl(struct super_block *sb, struct if_quotactl *qctl);
1123 extern const struct inode_operations ll_special_inode_operations;
1125 struct inode *ll_iget(struct super_block *sb, ino_t hash,
1126 struct lustre_md *lic);
1127 int ll_test_inode_by_fid(struct inode *inode, void *opaque);
1128 int ll_md_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1129 void *data, int flag);
1130 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de);
1131 int ll_rmdir_entry(struct inode *dir, char *name, int namelen);
1132 void ll_update_times(struct ptlrpc_request *request, struct inode *inode);
1135 int ll_writepage(struct page *page, struct writeback_control *wbc);
1136 int ll_writepages(struct address_space *, struct writeback_control *wbc);
1137 int ll_readpage(struct file *file, struct page *page);
1138 int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
1139 struct cl_page *page, struct file *file);
1140 void ll_readahead_init(struct inode *inode, struct ll_readahead_state *ras);
1141 int vvp_io_write_commit(const struct lu_env *env, struct cl_io *io);
1144 void ll_cl_add(struct inode *inode, const struct lu_env *env, struct cl_io *io,
1145 enum lcc_type type);
1146 void ll_cl_remove(struct inode *inode, const struct lu_env *env);
1147 struct ll_cl_context *ll_cl_find(struct inode *inode);
1149 extern const struct address_space_operations ll_aops;
1152 extern const struct inode_operations ll_file_inode_operations;
1153 const struct file_operations *ll_select_file_operations(struct ll_sb_info *sbi);
1154 extern int ll_have_md_lock(struct inode *inode, __u64 *bits,
1155 enum ldlm_mode l_req_mode);
1156 extern enum ldlm_mode ll_take_md_lock(struct inode *inode, __u64 bits,
1157 struct lustre_handle *lockh, __u64 flags,
1158 enum ldlm_mode mode);
1160 int ll_file_open(struct inode *inode, struct file *file);
1161 int ll_file_release(struct inode *inode, struct file *file);
1162 int ll_release_openhandle(struct dentry *, struct lookup_intent *);
1163 int ll_md_real_close(struct inode *inode, fmode_t fmode);
1164 void ll_track_file_opens(struct inode *inode);
1165 extern void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1166 struct ll_file_data *file, loff_t pos,
1167 size_t count, int rw);
1168 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
1169 int ll_getattr(struct user_namespace *mnt_userns, const struct path *path,
1170 struct kstat *stat, u32 request_mask, unsigned int flags);
1172 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
1173 #endif /* HAVE_USER_NAMESPACE_ARG */
1174 int ll_getattr_dentry(struct dentry *de, struct kstat *stat, u32 request_mask,
1175 unsigned int flags, bool foreign);
1176 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1177 struct posix_acl *ll_get_acl(struct inode *inode, int type
1178 #ifdef HAVE_GET_ACL_RCU_ARG
1180 #endif /* HAVE_GET_ACL_RCU_ARG */
1182 int ll_set_acl(struct user_namespace *mnt_userns, struct inode *inode,
1183 struct posix_acl *acl, int type);
1184 #else /* !CONFIG_LUSTRE_FS_POSIX_ACL */
1185 #define ll_get_acl NULL
1186 #define ll_set_acl NULL
1187 #endif /* CONFIG_LUSTRE_FS_POSIX_ACL */
1189 static inline int ll_xflags_to_inode_flags(int xflags)
1191 return ((xflags & FS_XFLAG_SYNC) ? S_SYNC : 0) |
1192 ((xflags & FS_XFLAG_NOATIME) ? S_NOATIME : 0) |
1193 ((xflags & FS_XFLAG_APPEND) ? S_APPEND : 0) |
1194 ((xflags & FS_XFLAG_IMMUTABLE) ? S_IMMUTABLE : 0);
1197 static inline int ll_inode_flags_to_xflags(int inode_flags)
1199 return ((inode_flags & S_SYNC) ? FS_XFLAG_SYNC : 0) |
1200 ((inode_flags & S_NOATIME) ? FS_XFLAG_NOATIME : 0) |
1201 ((inode_flags & S_APPEND) ? FS_XFLAG_APPEND : 0) |
1202 ((inode_flags & S_IMMUTABLE) ? FS_XFLAG_IMMUTABLE : 0);
1205 int ll_migrate(struct inode *parent, struct file *file,
1206 struct lmv_user_md *lum, const char *name, __u32 flags);
1207 int ll_get_fid_by_name(struct inode *parent, const char *name,
1208 int namelen, struct lu_fid *fid, struct inode **inode);
1209 int ll_inode_permission(struct user_namespace *mnt_userns, struct inode *inode,
1211 int ll_ioctl_check_project(struct inode *inode, __u32 xflags, __u32 projid);
1212 int ll_ioctl_fsgetxattr(struct inode *inode, unsigned int cmd,
1214 int ll_ioctl_fssetxattr(struct inode *inode, unsigned int cmd,
1216 int ll_ioctl_project(struct file *file, unsigned int cmd,
1219 int ll_lov_setstripe_ea_info(struct inode *inode, struct dentry *dentry,
1220 __u64 flags, struct lov_user_md *lum,
1222 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1223 struct lov_mds_md **lmm, int *lmm_size,
1224 struct ptlrpc_request **request);
1225 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
1227 int ll_dir_get_default_layout(struct inode *inode, void **plmm, int *plmm_size,
1228 struct ptlrpc_request **request, u64 valid,
1229 enum get_default_layout_type type);
1230 int ll_dir_getstripe_default(struct inode *inode, void **lmmp,
1231 int *lmm_size, struct ptlrpc_request **request,
1232 struct ptlrpc_request **root_request, u64 valid);
1233 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
1234 struct ptlrpc_request **request, u64 valid);
1235 int ll_fsync(struct file *file, loff_t start, loff_t end, int data);
1236 int ll_merge_attr(const struct lu_env *env, struct inode *inode);
1237 int ll_fid2path(struct inode *inode, void __user *arg);
1238 int ll_data_version(struct inode *inode, __u64 *data_version, int flags);
1239 int ll_hsm_release(struct inode *inode);
1240 int ll_hsm_state_set(struct inode *inode, struct hsm_state_set *hss);
1241 void ll_io_set_mirror(struct cl_io *io, const struct file *file);
1243 /* llite/dcache.c */
1245 extern const struct dentry_operations ll_d_ops;
1247 bool ll_d_setup(struct dentry *de, bool do_put);
1249 static inline bool lld_is_init(struct dentry *dentry)
1251 return ll_d2d(dentry);
1254 #define ll_d_setup(de, do_put) (true)
1255 #define lld_is_init(dentry) (true)
1258 void ll_intent_drop_lock(struct lookup_intent *);
1259 void ll_intent_release(struct lookup_intent *);
1260 void ll_prune_aliases(struct inode *inode);
1261 void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry);
1262 int ll_revalidate_it_finish(struct ptlrpc_request *request,
1263 struct lookup_intent *it, struct dentry *de);
1265 /* llite/llite_lib.c */
1266 extern const struct super_operations lustre_super_operations;
1268 void ll_lli_init(struct ll_inode_info *lli);
1269 int ll_fill_super(struct super_block *sb);
1270 void ll_put_super(struct super_block *sb);
1271 void ll_kill_super(struct super_block *sb);
1272 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock);
1273 void ll_dir_clear_lsm_md(struct inode *inode);
1274 void ll_clear_inode(struct inode *inode);
1275 int volatile_ref_file(const char *volatile_name, int volatile_len,
1276 struct file **ref_file);
1277 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1278 enum op_xvalid xvalid, bool hsm_import);
1279 int ll_setattr(struct user_namespace *mnt_userns, struct dentry *de,
1280 struct iattr *attr);
1281 int ll_statfs(struct dentry *de, struct kstatfs *sfs);
1282 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1284 int ll_update_inode(struct inode *inode, struct lustre_md *md);
1285 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags);
1286 void ll_update_dir_depth(struct inode *dir, struct inode *inode);
1287 int ll_read_inode2(struct inode *inode, void *opaque);
1288 void ll_truncate_inode_pages_final(struct inode *inode);
1289 void ll_delete_inode(struct inode *inode);
1290 int ll_iocontrol(struct inode *inode, struct file *file,
1291 unsigned int cmd, unsigned long arg);
1292 int ll_flush_ctx(struct inode *inode);
1293 void ll_umount_begin(struct super_block *sb);
1294 int ll_remount_fs(struct super_block *sb, int *flags, char *data);
1295 int ll_show_options(struct seq_file *seq, struct dentry *dentry);
1296 void ll_dirty_page_discard_warn(struct inode *inode, int ioret);
1297 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
1298 struct super_block *sb, struct lookup_intent *it);
1299 int ll_obd_statfs(struct inode *inode, void __user *arg);
1300 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *max_mdsize);
1301 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *default_mdsize);
1302 int ll_set_default_mdsize(struct ll_sb_info *sbi, int default_mdsize);
1304 void ll_unlock_md_op_lsm(struct md_op_data *op_data);
1305 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
1306 struct inode *i1, struct inode *i2,
1307 const char *name, size_t namelen,
1308 __u32 mode, enum md_op_code opc,
1310 void ll_finish_md_op_data(struct md_op_data *op_data);
1311 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg);
1312 void ll_compute_rootsquash_state(struct ll_sb_info *sbi);
1313 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
1314 struct lov_user_md **kbuf);
1315 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill);
1317 void ll_dom_finish_open(struct inode *inode, struct ptlrpc_request *req);
1319 /* Compute expected user md size when passing in a md from user space */
1320 static inline ssize_t ll_lov_user_md_size(const struct lov_user_md *lum)
1322 switch (lum->lmm_magic) {
1323 case LOV_USER_MAGIC_V1:
1324 return sizeof(struct lov_user_md_v1);
1325 case LOV_USER_MAGIC_V3:
1326 return sizeof(struct lov_user_md_v3);
1327 case LOV_USER_MAGIC_SPECIFIC:
1328 if (lum->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
1331 return lov_user_md_size(lum->lmm_stripe_count,
1332 LOV_USER_MAGIC_SPECIFIC);
1333 case LOV_USER_MAGIC_COMP_V1:
1334 return ((struct lov_comp_md_v1 *)lum)->lcm_size;
1335 case LOV_USER_MAGIC_FOREIGN:
1336 return foreign_size(lum);
1342 /* llite/llite_nfs.c */
1343 extern const struct export_operations lustre_export_operations;
1344 __u32 get_uuid2int(const char *name, int len);
1345 struct inode *search_inode_for_lustre(struct super_block *sb,
1346 const struct lu_fid *fid);
1347 int ll_dir_get_parent_fid(struct inode *dir, struct lu_fid *parent_fid);
1349 /* llite/symlink.c */
1350 extern const struct inode_operations ll_fast_symlink_inode_operations;
1353 * IO arguments for various VFS I/O interfaces.
1355 struct vvp_io_args {
1356 /** normal/sendfile/splice */
1359 struct kiocb *via_iocb;
1360 struct iov_iter *via_iter;
1370 struct ll_cl_context {
1371 struct list_head lcc_list;
1373 const struct lu_env *lcc_env;
1374 struct cl_io *lcc_io;
1375 struct cl_page *lcc_page;
1376 enum lcc_type lcc_type;
1379 struct ll_thread_info {
1380 struct vvp_io_args lti_args;
1381 struct ra_io_arg lti_ria;
1382 struct ll_cl_context lti_io_ctx;
1385 extern struct lu_context_key ll_thread_key;
1387 static inline struct ll_thread_info *ll_env_info(const struct lu_env *env)
1389 struct ll_thread_info *lti;
1391 lti = lu_context_key_get(&env->le_ctx, &ll_thread_key);
1392 LASSERT(lti != NULL);
1397 static inline struct vvp_io_args *ll_env_args(const struct lu_env *env)
1399 return &ll_env_info(env)->lti_args;
1402 void ll_io_init(struct cl_io *io, struct file *file, enum cl_io_type iot,
1403 struct vvp_io_args *args);
1405 /* llite/llite_mmap.c */
1407 int ll_file_mmap(struct file * file, struct vm_area_struct * vma);
1408 void policy_from_vma(union ldlm_policy_data *policy, struct vm_area_struct *vma,
1409 unsigned long addr, size_t count);
1410 struct vm_area_struct *our_vma(struct mm_struct *mm, unsigned long addr,
1413 #define ll_s2sbi(sb) (s2lsi(sb)->lsi_llsbi)
1415 /* don't need an addref as the sb_info should be holding one */
1416 static inline struct obd_export *ll_s2dtexp(struct super_block *sb)
1418 return ll_s2sbi(sb)->ll_dt_exp;
1421 /* don't need an addref as the sb_info should be holding one */
1422 static inline struct obd_export *ll_s2mdexp(struct super_block *sb)
1424 return ll_s2sbi(sb)->ll_md_exp;
1427 static inline struct client_obd *sbi2mdc(struct ll_sb_info *sbi)
1429 struct obd_device *obd = sbi->ll_md_exp->exp_obd;
1435 // FIXME: replace the name of this with LL_SB to conform to kernel stuff
1436 static inline struct ll_sb_info *ll_i2sbi(struct inode *inode)
1438 return ll_s2sbi(inode->i_sb);
1441 static inline struct obd_export *ll_i2dtexp(struct inode *inode)
1443 return ll_s2dtexp(inode->i_sb);
1446 static inline struct obd_export *ll_i2mdexp(struct inode *inode)
1448 return ll_s2mdexp(inode->i_sb);
1451 static inline struct lu_fid *ll_inode2fid(struct inode *inode)
1455 LASSERT(inode != NULL);
1456 fid = &ll_i2info(inode)->lli_fid;
1461 static inline bool ll_dir_striped(struct inode *inode)
1463 struct ll_inode_info *lli;
1467 if (!S_ISDIR(inode->i_mode))
1470 lli = ll_i2info(inode);
1471 if (!lli->lli_lsm_md)
1474 down_read(&lli->lli_lsm_sem);
1475 rc = lmv_dir_striped(lli->lli_lsm_md);
1476 up_read(&lli->lli_lsm_sem);
1481 static inline loff_t ll_file_maxbytes(struct inode *inode)
1483 struct cl_object *obj = ll_i2info(inode)->lli_clob;
1486 return MAX_LFS_FILESIZE;
1488 return min_t(loff_t, cl_object_maxbytes(obj), MAX_LFS_FILESIZE);
1492 extern const struct xattr_handler *ll_xattr_handlers[];
1494 #define XATTR_USER_T 1
1495 #define XATTR_TRUSTED_T 2
1496 #define XATTR_SECURITY_T 3
1497 #define XATTR_ACL_ACCESS_T 4
1498 #define XATTR_ACL_DEFAULT_T 5
1499 #define XATTR_LUSTRE_T 6
1500 #define XATTR_OTHER_T 7
1501 #define XATTR_ENCRYPTION_T 9
1503 ssize_t ll_listxattr(struct dentry *dentry, char *buffer, size_t size);
1504 int ll_xattr_list(struct inode *inode, const char *name, int type,
1505 void *buffer, size_t size, u64 valid);
1506 const struct xattr_handler *get_xattr_type(const char *name);
1509 * Common IO arguments for various VFS I/O interfaces.
1511 int cl_sb_init(struct super_block *sb);
1512 int cl_sb_fini(struct super_block *sb);
1514 enum ras_update_flags {
1518 void ll_ra_count_put(struct ll_sb_info *sbi, unsigned long len);
1519 void ll_ra_stats_inc(struct inode *inode, enum ra_stat which);
1523 #define LL_SA_RPC_MIN 2
1524 #define LL_SA_RPC_DEF 32
1525 #define LL_SA_RPC_MAX 512
1527 /* XXX: If want to support more concurrent statahead instances,
1528 * please consider to decentralize the RPC lists attached
1529 * on related import, such as imp_{sending,delayed}_list.
1531 #define LL_SA_RUNNING_MAX 256
1532 #define LL_SA_RUNNING_DEF 16
1534 #define LL_SA_CACHE_BIT 5
1535 #define LL_SA_CACHE_SIZE (1 << LL_SA_CACHE_BIT)
1536 #define LL_SA_CACHE_MASK (LL_SA_CACHE_SIZE - 1)
1538 /* per inode struct, for dir only */
1539 struct ll_statahead_info {
1540 struct dentry *sai_dentry;
1541 atomic_t sai_refcount; /* when access this struct, hold
1543 unsigned int sai_max; /* max ahead of lookup */
1544 __u64 sai_sent; /* stat requests sent count */
1545 __u64 sai_replied; /* stat requests which received
1547 __u64 sai_index; /* index of statahead entry */
1548 __u64 sai_index_wait; /* index of entry which is the
1549 * caller is waiting for */
1550 __u64 sai_hit; /* hit count */
1551 __u64 sai_miss; /* miss count:
1552 * for "ls -al" case, includes
1553 * hidden dentry miss;
1554 * for "ls -l" case, it does not
1555 * include hidden dentry miss.
1556 * "sai_miss_hidden" is used for
1559 unsigned int sai_consecutive_miss; /* consecutive miss */
1560 unsigned int sai_miss_hidden;/* "ls -al", but first dentry
1561 * is not a hidden one */
1562 unsigned int sai_skip_hidden;/* skipped hidden dentry count
1564 unsigned int sai_ls_all:1; /* "ls -al", do stat-ahead for
1566 wait_queue_head_t sai_waitq; /* stat-ahead wait queue */
1567 struct task_struct *sai_task; /* stat-ahead thread */
1568 struct task_struct *sai_agl_task; /* AGL thread */
1569 struct list_head sai_entries; /* completed entries */
1570 struct list_head sai_agls; /* AGLs to be sent */
1571 struct list_head sai_cache[LL_SA_CACHE_SIZE];
1572 spinlock_t sai_cache_lock[LL_SA_CACHE_SIZE];
1573 atomic_t sai_cache_count; /* entry count in cache */
1576 int ll_revalidate_statahead(struct inode *dir, struct dentry **dentry,
1578 int ll_start_statahead(struct inode *dir, struct dentry *dentry, bool agl);
1579 void ll_authorize_statahead(struct inode *dir, void *key);
1580 void ll_deauthorize_statahead(struct inode *dir, void *key);
1583 blkcnt_t dirty_cnt(struct inode *inode);
1585 int cl_glimpse_size0(struct inode *inode, int agl);
1586 int cl_glimpse_lock(const struct lu_env *env, struct cl_io *io,
1587 struct inode *inode, struct cl_object *clob, int agl);
1589 static inline int cl_glimpse_size(struct inode *inode)
1591 return cl_glimpse_size0(inode, 0);
1594 /* AGL is 'asychronous glimpse lock', which is a speculative lock taken as
1595 * part of statahead */
1596 static inline int cl_agl(struct inode *inode)
1598 return cl_glimpse_size0(inode, 1);
1601 int ll_file_lock_ahead(struct file *file, struct llapi_lu_ladvise *ladvise);
1603 int cl_io_get(struct inode *inode, struct lu_env **envout,
1604 struct cl_io **ioout, __u16 *refcheck);
1606 static inline int ll_glimpse_size(struct inode *inode)
1608 struct ll_inode_info *lli = ll_i2info(inode);
1611 down_read(&lli->lli_glimpse_sem);
1612 rc = cl_glimpse_size(inode);
1613 lli->lli_glimpse_time = ktime_get();
1614 up_read(&lli->lli_glimpse_sem);
1618 /* dentry may statahead when statahead is enabled and current process has opened
1619 * parent directory, and this dentry hasn't accessed statahead cache before */
1621 dentry_may_statahead(struct inode *dir, struct dentry *dentry)
1623 struct ll_inode_info *lli;
1624 struct ll_dentry_data *ldd;
1626 if (ll_i2sbi(dir)->ll_sa_max == 0)
1629 lli = ll_i2info(dir);
1631 /* statahead is not allowed for this dir, there may be three causes:
1632 * 1. dir is not opened.
1633 * 2. statahead hit ratio is too low.
1634 * 3. previous stat started statahead thread failed. */
1635 if (!lli->lli_sa_enabled)
1638 /* not the same process, don't statahead */
1639 if (lli->lli_opendir_pid != current->pid)
1643 * When stating a dentry, kernel may trigger 'revalidate' or 'lookup'
1644 * multiple times, eg. for 'getattr', 'getxattr' and etc.
1645 * For patchless client, lookup intent is not accurate, which may
1646 * misguide statahead. For example:
1647 * The 'revalidate' call for 'getattr' and 'getxattr' of a dentry will
1648 * have the same intent -- IT_GETATTR, while one dentry should access
1649 * statahead cache once, otherwise statahead windows is messed up.
1650 * The solution is as following:
1651 * Assign 'lld_sa_generation' with 'lli_sa_generation' when a dentry
1652 * IT_GETATTR for the first time, and subsequent IT_GETATTR will
1653 * bypass interacting with statahead cache by checking
1654 * 'lld_sa_generation == lli->lli_sa_generation'.
1656 ldd = ll_d2d(dentry);
1657 if (ldd != NULL && lli->lli_sa_generation &&
1658 ldd->lld_sa_generation == lli->lli_sa_generation)
1664 int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end,
1665 enum cl_fsync_mode mode, int ignore_layout);
1667 static inline int ll_file_nolock(const struct file *file)
1669 struct ll_file_data *fd = file->private_data;
1670 struct inode *inode = file_inode((struct file *)file);
1672 LASSERT(fd != NULL);
1673 return ((fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
1674 test_bit(LL_SBI_NOLCK, ll_i2sbi(inode)->ll_flags));
1677 static inline void ll_set_lock_data(struct obd_export *exp, struct inode *inode,
1678 struct lookup_intent *it, __u64 *bits)
1680 if (!it->it_lock_set) {
1681 struct lustre_handle handle;
1683 /* If this inode is a remote object, it will get two
1684 * separate locks in different namespaces, Master MDT,
1685 * where the name entry is, will grant LOOKUP lock,
1686 * remote MDT, where the object is, will grant
1687 * UPDATE|PERM lock. The inode will be attched to both
1688 * LOOKUP and PERM locks, so revoking either locks will
1689 * case the dcache being cleared */
1690 if (it->it_remote_lock_mode) {
1691 handle.cookie = it->it_remote_lock_handle;
1692 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID
1693 "(%p) for remote lock %#llx\n",
1694 PFID(ll_inode2fid(inode)), inode,
1696 md_set_lock_data(exp, &handle, inode, NULL);
1699 handle.cookie = it->it_lock_handle;
1701 CDEBUG(D_DLMTRACE, "setting l_data to inode "DFID"(%p)"
1702 " for lock %#llx\n",
1703 PFID(ll_inode2fid(inode)), inode, handle.cookie);
1705 md_set_lock_data(exp, &handle, inode, &it->it_lock_bits);
1706 it->it_lock_set = 1;
1710 *bits = it->it_lock_bits;
1713 static inline int d_lustre_invalid(const struct dentry *dentry)
1715 return !ll_d2d(dentry) || ll_d2d(dentry)->lld_invalid;
1719 * Mark dentry INVALID, if dentry refcount is zero (this is normally case for
1720 * ll_md_blocking_ast), it will be pruned by ll_prune_aliases() and
1721 * ll_prune_negative_children(); otherwise dput() of the last refcount will
1722 * unhash this dentry and kill it.
1724 static inline void d_lustre_invalidate(struct dentry *dentry)
1726 CDEBUG(D_DENTRY, "invalidate dentry %pd (%p) parent %p inode %p refc %d\n",
1728 dentry->d_parent, dentry->d_inode, ll_d_count(dentry));
1730 spin_lock(&dentry->d_lock);
1731 if (lld_is_init(dentry))
1732 ll_d2d(dentry)->lld_invalid = 1;
1733 spin_unlock(&dentry->d_lock);
1736 static inline void d_lustre_revalidate(struct dentry *dentry)
1738 spin_lock(&dentry->d_lock);
1739 LASSERT(ll_d2d(dentry));
1740 ll_d2d(dentry)->lld_invalid = 0;
1741 spin_unlock(&dentry->d_lock);
1744 static inline dev_t ll_compat_encode_dev(dev_t dev)
1746 /* The compat_sys_*stat*() syscalls will fail unless the
1747 * device majors and minors are both less than 256. Note that
1748 * the value returned here will be passed through
1749 * old_encode_dev() in cp_compat_stat(). And so we are not
1750 * trying to return a valid compat (u16) device number, just
1751 * one that will pass the old_valid_dev() check. */
1753 return MKDEV(MAJOR(dev) & 0xff, MINOR(dev) & 0xff);
1756 int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf);
1757 int ll_layout_refresh(struct inode *inode, __u32 *gen);
1758 int ll_layout_restore(struct inode *inode, loff_t start, __u64 length);
1759 int ll_layout_write_intent(struct inode *inode, enum layout_intent_opc opc,
1760 struct lu_extent *ext);
1762 int ll_xattr_init(void);
1763 void ll_xattr_fini(void);
1765 int ll_page_sync_io(const struct lu_env *env, struct cl_io *io,
1766 struct cl_page *page, enum cl_req_type crt);
1768 int ll_getparent(struct file *file, struct getparent __user *arg);
1771 int cl_setattr_ost(struct cl_object *obj, const struct iattr *attr,
1772 enum op_xvalid xvalid, unsigned int attr_flags);
1774 extern struct lu_env *cl_inode_fini_env;
1775 extern __u16 cl_inode_fini_refcheck;
1777 int cl_file_inode_init(struct inode *inode, struct lustre_md *md);
1778 void cl_inode_fini(struct inode *inode);
1780 u64 cl_fid_build_ino(const struct lu_fid *fid, int api32);
1781 u32 cl_fid_build_gen(const struct lu_fid *fid);
1783 static inline struct pcc_super *ll_i2pccs(struct inode *inode)
1785 return &ll_i2sbi(inode)->ll_pcc_super;
1788 static inline struct pcc_super *ll_info2pccs(struct ll_inode_info *lli)
1790 return ll_i2pccs(ll_info2i(lli));
1794 /* The digested form is made of a FID (16 bytes) followed by the second-to-last
1795 * ciphertext block (16 bytes), so a total length of 32 bytes.
1796 * That way, llcrypt does not compute a digested form of this digest.
1798 struct ll_digest_filename {
1799 struct lu_fid ldf_fid;
1800 char ldf_excerpt[LL_CRYPTO_BLOCK_SIZE];
1803 int ll_setup_filename(struct inode *dir, const struct qstr *iname,
1804 int lookup, struct llcrypt_name *fname,
1805 struct lu_fid *fid);
1806 int ll_fname_disk_to_usr(struct inode *inode,
1807 u32 hash, u32 minor_hash,
1808 struct llcrypt_str *iname, struct llcrypt_str *oname,
1809 struct lu_fid *fid);
1810 int ll_revalidate_d_crypto(struct dentry *dentry, unsigned int flags);
1811 int ll_file_open_encrypt(struct inode *inode, struct file *filp);
1812 static inline char *xattr_for_enc(struct inode *inode)
1814 if (ll_sbi_has_name_encrypt(ll_i2sbi(inode)))
1815 return LL_XATTR_NAME_ENCRYPTION_CONTEXT;
1817 return LL_XATTR_NAME_ENCRYPTION_CONTEXT_OLD;
1819 #ifdef HAVE_LUSTRE_CRYPTO
1820 extern const struct llcrypt_operations lustre_cryptops;
1823 /* llite/llite_foreign.c */
1824 int ll_manage_foreign(struct inode *inode, struct lustre_md *lmd);
1825 bool ll_foreign_is_openable(struct dentry *dentry, unsigned int flags);
1826 bool ll_foreign_is_removable(struct dentry *dentry, bool unset);
1828 int ll_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf);
1830 #endif /* LLITE_INTERNAL_H */