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) 2017, Commissariat a l'Energie Atomique et aux Energies
26 * Author: Henri Doreau <henri.doreau@cea.fr>
29 #define DEBUG_SUBSYSTEM S_MDC
31 #include <linux/init.h>
32 #include <linux/kthread.h>
33 #include <linux/poll.h>
34 #include <linux/miscdevice.h>
36 #include <lustre_log.h>
38 #include "mdc_internal.h"
42 * -- Changelog delivery through character device --
46 * Mutex to protect chlg_registered_devices below
48 static DEFINE_MUTEX(chlg_registered_dev_lock);
51 * Global linked list of all registered devices (one per MDT).
53 static LIST_HEAD(chlg_registered_devices);
56 struct chlg_registered_dev {
57 /* Device name of the form "changelog-{MDTNAME}" */
59 /* Misc device descriptor */
60 struct miscdevice ced_misc;
61 /* OBDs referencing this device (multiple mount point) */
62 struct list_head ced_obds;
63 /* Reference counter for proper deregistration */
65 /* Link within the global chlg_registered_devices */
66 struct list_head ced_link;
69 struct chlg_reader_state {
70 /* Shortcut to the corresponding OBD device */
71 struct obd_device *crs_obd;
72 /* Producer thread (if any) */
73 struct task_struct *crs_prod_task;
74 /* An error occurred that prevents from reading further */
76 /* EOF, no more records available */
78 /* Desired start position */
79 __u64 crs_start_offset;
80 /* Wait queue for the catalog processing thread */
81 wait_queue_head_t crs_waitq_prod;
82 /* Wait queue for the record copy threads */
83 wait_queue_head_t crs_waitq_cons;
84 /* Mutex protecting crs_rec_count and crs_rec_queue */
85 struct mutex crs_lock;
86 /* Number of item in the list */
88 /* List of prefetched enqueued_record::enq_linkage_items */
89 struct list_head crs_rec_queue;
92 struct chlg_rec_entry {
93 /* Link within the chlg_reader_state::crs_rec_queue list */
94 struct list_head enq_linkage;
95 /* Data (enq_record) field length */
97 /* Copy of a changelog record (see struct llog_changelog_rec) */
98 struct changelog_rec enq_record[];
102 /* Number of records to prefetch locally. */
103 CDEV_CHLG_MAX_PREFETCH = 1024,
107 * ChangeLog catalog processing callback invoked on each record.
108 * If the current record is eligible to userland delivery, push
109 * it into the crs_rec_queue where the consumer code will fetch it.
111 * @param[in] env (unused)
112 * @param[in] llh Client-side handle used to identify the llog
113 * @param[in] hdr Header of the current llog record
114 * @param[in,out] data chlg_reader_state passed from caller
116 * @return 0 or LLOG_PROC_* control code on success, negated error on failure.
118 static int chlg_read_cat_process_cb(const struct lu_env *env,
119 struct llog_handle *llh,
120 struct llog_rec_hdr *hdr, void *data)
122 struct llog_changelog_rec *rec;
123 struct chlg_reader_state *crs = data;
124 struct chlg_rec_entry *enq;
125 struct l_wait_info lwi = { 0 };
130 LASSERT(crs != NULL);
131 LASSERT(hdr != NULL);
133 rec = container_of(hdr, struct llog_changelog_rec, cr_hdr);
135 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
137 CERROR("%s: not a changelog rec %x/%d in llog "DFID" rc = %d\n",
138 crs->crs_obd->obd_name, rec->cr_hdr.lrh_type,
140 PFID(lu_object_fid(&llh->lgh_obj->do_lu)), rc);
144 /* Skip undesired records */
145 if (rec->cr.cr_index < crs->crs_start_offset)
148 CDEBUG(D_HSM, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID" %.*s\n",
149 rec->cr.cr_index, rec->cr.cr_type,
150 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
151 rec->cr.cr_flags & CLF_FLAGMASK,
152 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
153 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
155 l_wait_event(crs->crs_waitq_prod,
156 (crs->crs_rec_count < CDEV_CHLG_MAX_PREFETCH ||
157 kthread_should_stop()), &lwi);
159 if (kthread_should_stop())
160 RETURN(LLOG_PROC_BREAK);
162 len = changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
163 OBD_ALLOC(enq, sizeof(*enq) + len);
167 INIT_LIST_HEAD(&enq->enq_linkage);
168 enq->enq_length = len;
169 memcpy(enq->enq_record, &rec->cr, len);
171 mutex_lock(&crs->crs_lock);
172 list_add_tail(&enq->enq_linkage, &crs->crs_rec_queue);
173 crs->crs_rec_count++;
174 mutex_unlock(&crs->crs_lock);
176 wake_up_all(&crs->crs_waitq_cons);
182 * Remove record from the list it is attached to and free it.
184 static void enq_record_delete(struct chlg_rec_entry *rec)
186 list_del(&rec->enq_linkage);
187 OBD_FREE(rec, sizeof(*rec) + rec->enq_length);
191 * Record prefetch thread entry point. Opens the changelog catalog and starts
194 * @param[in,out] args chlg_reader_state passed from caller.
195 * @return 0 on success, negated error code on failure.
197 static int chlg_load(void *args)
199 struct chlg_reader_state *crs = args;
200 struct obd_device *obd = crs->crs_obd;
201 struct llog_ctxt *ctx = NULL;
202 struct llog_handle *llh = NULL;
203 struct l_wait_info lwi = { 0 };
207 ctx = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
209 GOTO(err_out, rc = -ENOENT);
211 rc = llog_open(NULL, ctx, &llh, NULL, CHANGELOG_CATALOG,
214 CERROR("%s: fail to open changelog catalog: rc = %d\n",
219 rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT|LLOG_F_EXT_JOBID, NULL);
221 CERROR("%s: fail to init llog handle: rc = %d\n",
226 rc = llog_cat_process(NULL, llh, chlg_read_cat_process_cb, crs, 0, 0);
228 CERROR("%s: fail to process llog: rc = %d\n", obd->obd_name, rc);
238 wake_up_all(&crs->crs_waitq_cons);
241 llog_cat_close(NULL, llh);
246 l_wait_event(crs->crs_waitq_prod, kthread_should_stop(), &lwi);
252 * Read handler, dequeues records from the chlg_reader_state if any.
253 * No partial records are copied to userland so this function can return less
254 * data than required (short read).
256 * @param[in] file File pointer to the character device.
257 * @param[out] buff Userland buffer where to copy the records.
258 * @param[in] count Userland buffer size.
259 * @param[out] ppos File position, updated with the index number of the next
261 * @return number of copied bytes on success, negated error code on failure.
263 static ssize_t chlg_read(struct file *file, char __user *buff, size_t count,
266 struct chlg_reader_state *crs = file->private_data;
267 struct chlg_rec_entry *rec;
268 struct chlg_rec_entry *tmp;
269 struct l_wait_info lwi = { 0 };
270 ssize_t written_total = 0;
274 if (file->f_flags & O_NONBLOCK && crs->crs_rec_count == 0)
277 l_wait_event(crs->crs_waitq_cons,
278 crs->crs_rec_count > 0 || crs->crs_eof || crs->crs_err,
281 mutex_lock(&crs->crs_lock);
282 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage) {
283 if (written_total + rec->enq_length > count)
286 if (copy_to_user(buff, rec->enq_record, rec->enq_length)) {
287 if (written_total == 0)
288 written_total = -EFAULT;
292 buff += rec->enq_length;
293 written_total += rec->enq_length;
295 crs->crs_rec_count--;
296 list_move_tail(&rec->enq_linkage, &consumed);
298 crs->crs_start_offset = rec->enq_record->cr_index + 1;
300 mutex_unlock(&crs->crs_lock);
302 if (written_total > 0)
303 wake_up_all(&crs->crs_waitq_prod);
305 list_for_each_entry_safe(rec, tmp, &consumed, enq_linkage)
306 enq_record_delete(rec);
308 *ppos = crs->crs_start_offset;
310 RETURN(written_total);
314 * Jump to a given record index. Helper for chlg_llseek().
316 * @param[in,out] crs Internal reader state.
317 * @param[in] offset Desired offset (index record).
318 * @return 0 on success, negated error code on failure.
320 static int chlg_set_start_offset(struct chlg_reader_state *crs, __u64 offset)
322 struct chlg_rec_entry *rec;
323 struct chlg_rec_entry *tmp;
325 mutex_lock(&crs->crs_lock);
326 if (offset < crs->crs_start_offset) {
327 mutex_unlock(&crs->crs_lock);
331 crs->crs_start_offset = offset;
332 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage) {
333 struct changelog_rec *cr = rec->enq_record;
335 if (cr->cr_index >= crs->crs_start_offset)
338 crs->crs_rec_count--;
339 enq_record_delete(rec);
342 mutex_unlock(&crs->crs_lock);
343 wake_up_all(&crs->crs_waitq_prod);
348 * Move read pointer to a certain record index, encoded as an offset.
350 * @param[in,out] file File pointer to the changelog character device
351 * @param[in] off Offset to skip, actually a record index, not byte count
352 * @param[in] whence Relative/Absolute interpretation of the offset
353 * @return the resulting position on success or negated error code on failure.
355 static loff_t chlg_llseek(struct file *file, loff_t off, int whence)
357 struct chlg_reader_state *crs = file->private_data;
366 pos = file->f_pos + off;
373 /* We cannot go backward */
374 if (pos < file->f_pos)
377 rc = chlg_set_start_offset(crs, pos);
386 * Clear record range for a given changelog reader.
388 * @param[in] crs Current internal state.
389 * @param[in] reader Changelog reader ID (cl1, cl2...)
390 * @param[in] record Record index up which to clear
391 * @return 0 on success, negated error code on failure.
393 static int chlg_clear(struct chlg_reader_state *crs, __u32 reader, __u64 record)
395 struct obd_device *obd = crs->crs_obd;
396 struct changelog_setinfo cs = {
401 return obd_set_info_async(NULL, obd->obd_self_export,
402 strlen(KEY_CHANGELOG_CLEAR),
403 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs, NULL);
406 /** Maximum changelog control command size */
407 #define CHLG_CONTROL_CMD_MAX 64
410 * Handle writes() into the changelog character device. Write() can be used
411 * to request special control operations.
413 * @param[in] file File pointer to the changelog character device
414 * @param[in] buff User supplied data (written data)
415 * @param[in] count Number of written bytes
416 * @param[in] off (unused)
417 * @return number of written bytes on success, negated error code on failure.
419 static ssize_t chlg_write(struct file *file, const char __user *buff,
420 size_t count, loff_t *off)
422 struct chlg_reader_state *crs = file->private_data;
429 if (count > CHLG_CONTROL_CMD_MAX)
432 OBD_ALLOC(kbuf, CHLG_CONTROL_CMD_MAX);
436 if (copy_from_user(kbuf, buff, count))
437 GOTO(out_kbuf, rc = -EFAULT);
439 kbuf[CHLG_CONTROL_CMD_MAX - 1] = '\0';
441 if (sscanf(kbuf, "clear:cl%u:%llu", &reader, &record) == 2)
442 rc = chlg_clear(crs, reader, record);
448 OBD_FREE(kbuf, CHLG_CONTROL_CMD_MAX);
449 return rc < 0 ? rc : count;
453 * Find the OBD device associated to a changelog character device.
454 * @param[in] cdev character device instance descriptor
455 * @return corresponding OBD device or NULL if none was found.
457 static struct obd_device *chlg_obd_get(dev_t cdev)
459 int minor = MINOR(cdev);
460 struct obd_device *obd = NULL;
461 struct chlg_registered_dev *curr;
463 mutex_lock(&chlg_registered_dev_lock);
464 list_for_each_entry(curr, &chlg_registered_devices, ced_link) {
465 if (curr->ced_misc.minor == minor) {
466 /* take the first available OBD device attached */
467 obd = list_first_entry(&curr->ced_obds,
469 u.cli.cl_chg_dev_linkage);
473 mutex_unlock(&chlg_registered_dev_lock);
478 * Open handler, initialize internal CRS state and spawn prefetch thread if
480 * @param[in] inode Inode struct for the open character device.
481 * @param[in] file Corresponding file pointer.
482 * @return 0 on success, negated error code on failure.
484 static int chlg_open(struct inode *inode, struct file *file)
486 struct chlg_reader_state *crs;
487 struct obd_device *obd = chlg_obd_get(inode->i_rdev);
488 struct task_struct *task;
500 crs->crs_err = false;
501 crs->crs_eof = false;
503 mutex_init(&crs->crs_lock);
504 INIT_LIST_HEAD(&crs->crs_rec_queue);
505 init_waitqueue_head(&crs->crs_waitq_prod);
506 init_waitqueue_head(&crs->crs_waitq_cons);
508 if (file->f_mode & FMODE_READ) {
509 task = kthread_run(chlg_load, crs, "chlg_load_thread");
512 CERROR("%s: cannot start changelog thread: rc = %d\n",
516 crs->crs_prod_task = task;
519 file->private_data = crs;
528 * Close handler, release resources.
530 * @param[in] inode Inode struct for the open character device.
531 * @param[in] file Corresponding file pointer.
532 * @return 0 on success, negated error code on failure.
534 static int chlg_release(struct inode *inode, struct file *file)
536 struct chlg_reader_state *crs = file->private_data;
537 struct chlg_rec_entry *rec;
538 struct chlg_rec_entry *tmp;
540 if (crs->crs_prod_task)
541 kthread_stop(crs->crs_prod_task);
543 list_for_each_entry_safe(rec, tmp, &crs->crs_rec_queue, enq_linkage)
544 enq_record_delete(rec);
551 * Poll handler, indicates whether the device is readable (new records) and
554 * @param[in] file Device file pointer.
555 * @param[in] wait (opaque)
556 * @return combination of the poll status flags.
558 static unsigned int chlg_poll(struct file *file, poll_table *wait)
560 struct chlg_reader_state *crs = file->private_data;
561 unsigned int mask = 0;
563 mutex_lock(&crs->crs_lock);
564 poll_wait(file, &crs->crs_waitq_cons, wait);
565 if (crs->crs_rec_count > 0)
566 mask |= POLLIN | POLLRDNORM;
571 mutex_unlock(&crs->crs_lock);
575 static const struct file_operations chlg_fops = {
576 .owner = THIS_MODULE,
577 .llseek = chlg_llseek,
581 .release = chlg_release,
586 * This uses obd_name of the form: "testfs-MDT0000-mdc-ffff88006501600"
587 * and returns a name of the form: "changelog-testfs-MDT0000".
589 static void get_chlg_name(char *name, size_t name_len, struct obd_device *obd)
593 snprintf(name, name_len, "changelog-%s", obd->obd_name);
595 /* Find the 2nd '-' from the end and truncate on it */
596 for (i = 0; i < 2; i++) {
597 char *p = strrchr(name, '-');
606 * Find a changelog character device by name.
607 * All devices registered during MDC setup are listed in a global list with
608 * their names attached.
610 static struct chlg_registered_dev *
611 chlg_registered_dev_find_by_name(const char *name)
613 struct chlg_registered_dev *dit;
615 list_for_each_entry(dit, &chlg_registered_devices, ced_link)
616 if (strcmp(name, dit->ced_name) == 0)
622 * Find chlg_registered_dev structure for a given OBD device.
623 * This is bad O(n^2) but for each filesystem:
624 * - N is # of MDTs times # of mount points
625 * - this only runs at shutdown
627 static struct chlg_registered_dev *
628 chlg_registered_dev_find_by_obd(const struct obd_device *obd)
630 struct chlg_registered_dev *dit;
631 struct obd_device *oit;
633 list_for_each_entry(dit, &chlg_registered_devices, ced_link)
634 list_for_each_entry(oit, &dit->ced_obds,
635 u.cli.cl_chg_dev_linkage)
642 * Changelog character device initialization.
643 * Register a misc character device with a dynamic minor number, under a name
644 * of the form: 'changelog-fsname-MDTxxxx'. Reference this OBD device with it.
646 * @param[in] obd This MDC obd_device.
647 * @return 0 on success, negated error code on failure.
649 int mdc_changelog_cdev_init(struct obd_device *obd)
651 struct chlg_registered_dev *exist;
652 struct chlg_registered_dev *entry;
656 OBD_ALLOC_PTR(entry);
660 get_chlg_name(entry->ced_name, sizeof(entry->ced_name), obd);
662 entry->ced_misc.minor = MISC_DYNAMIC_MINOR;
663 entry->ced_misc.name = entry->ced_name;
664 entry->ced_misc.fops = &chlg_fops;
666 kref_init(&entry->ced_refs);
667 INIT_LIST_HEAD(&entry->ced_obds);
668 INIT_LIST_HEAD(&entry->ced_link);
670 mutex_lock(&chlg_registered_dev_lock);
671 exist = chlg_registered_dev_find_by_name(entry->ced_name);
673 kref_get(&exist->ced_refs);
674 list_add_tail(&obd->u.cli.cl_chg_dev_linkage, &exist->ced_obds);
675 GOTO(out_unlock, rc = 0);
678 /* Register new character device */
679 rc = misc_register(&entry->ced_misc);
681 GOTO(out_unlock, rc);
683 list_add_tail(&obd->u.cli.cl_chg_dev_linkage, &entry->ced_obds);
684 list_add_tail(&entry->ced_link, &chlg_registered_devices);
686 entry = NULL; /* prevent it from being freed below */
689 mutex_unlock(&chlg_registered_dev_lock);
696 * Deregister a changelog character device whose refcount has reached zero.
698 static void chlg_dev_clear(struct kref *kref)
700 struct chlg_registered_dev *entry = container_of(kref,
701 struct chlg_registered_dev,
705 list_del(&entry->ced_link);
706 misc_deregister(&entry->ced_misc);
712 * Release OBD, decrease reference count of the corresponding changelog device.
714 void mdc_changelog_cdev_finish(struct obd_device *obd)
716 struct chlg_registered_dev *dev = chlg_registered_dev_find_by_obd(obd);
719 mutex_lock(&chlg_registered_dev_lock);
720 list_del_init(&obd->u.cli.cl_chg_dev_linkage);
721 kref_put(&dev->ced_refs, chlg_dev_clear);
722 mutex_unlock(&chlg_registered_dev_lock);