brintos

brintos / linux-shallow public Read only

0
0
Text · 35.3 KiB · 694db61 Raw
1470 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 * Provide a pstore intermediate backend, organized into kernel memory4 * allocated zones that are then mapped and flushed into a single5 * contiguous region on a storage backend of some kind (block, mtd, etc).6 */7 8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt9 10#include <linux/kernel.h>11#include <linux/module.h>12#include <linux/slab.h>13#include <linux/mount.h>14#include <linux/printk.h>15#include <linux/fs.h>16#include <linux/pstore_zone.h>17#include <linux/kdev_t.h>18#include <linux/device.h>19#include <linux/namei.h>20#include <linux/fcntl.h>21#include <linux/uio.h>22#include <linux/writeback.h>23#include "internal.h"24 25/**26 * struct psz_buffer - header of zone to flush to storage27 *28 * @sig: signature to indicate header (PSZ_SIG xor PSZONE-type value)29 * @datalen: length of data in @data30 * @start: offset into @data where the beginning of the stored bytes begin31 * @data: zone data.32 */33struct psz_buffer {34#define PSZ_SIG (0x43474244) /* DBGC */35	uint32_t sig;36	atomic_t datalen;37	atomic_t start;38	uint8_t data[];39};40 41/**42 * struct psz_kmsg_header - kmsg dump-specific header to flush to storage43 *44 * @magic: magic num for kmsg dump header45 * @time: kmsg dump trigger time46 * @compressed: whether conpressed47 * @counter: kmsg dump counter48 * @reason: the kmsg dump reason (e.g. oops, panic, etc)49 * @data: pointer to log data50 *51 * This is a sub-header for a kmsg dump, trailing after &psz_buffer.52 */53struct psz_kmsg_header {54#define PSTORE_KMSG_HEADER_MAGIC 0x4dfc3ae5 /* Just a random number */55	uint32_t magic;56	struct timespec64 time;57	bool compressed;58	uint32_t counter;59	enum kmsg_dump_reason reason;60	uint8_t data[];61};62 63/**64 * struct pstore_zone - single stored buffer65 *66 * @off: zone offset of storage67 * @type: front-end type for this zone68 * @name: front-end name for this zone69 * @buffer: pointer to data buffer managed by this zone70 * @oldbuf: pointer to old data buffer71 * @buffer_size: bytes in @buffer->data72 * @should_recover: whether this zone should recover from storage73 * @dirty: whether the data in @buffer dirty74 *75 * zone structure in memory.76 */77struct pstore_zone {78	loff_t off;79	const char *name;80	enum pstore_type_id type;81 82	struct psz_buffer *buffer;83	struct psz_buffer *oldbuf;84	size_t buffer_size;85	bool should_recover;86	atomic_t dirty;87};88 89/**90 * struct psz_context - all about running state of pstore/zone91 *92 * @kpszs: kmsg dump storage zones93 * @ppsz: pmsg storage zone94 * @cpsz: console storage zone95 * @fpszs: ftrace storage zones96 * @kmsg_max_cnt: max count of @kpszs97 * @kmsg_read_cnt: counter of total read kmsg dumps98 * @kmsg_write_cnt: counter of total kmsg dump writes99 * @pmsg_read_cnt: counter of total read pmsg zone100 * @console_read_cnt: counter of total read console zone101 * @ftrace_max_cnt: max count of @fpszs102 * @ftrace_read_cnt: counter of max read ftrace zone103 * @oops_counter: counter of oops dumps104 * @panic_counter: counter of panic dumps105 * @recovered: whether finished recovering data from storage106 * @on_panic: whether panic is happening107 * @pstore_zone_info_lock: lock to @pstore_zone_info108 * @pstore_zone_info: information from backend109 * @pstore: structure for pstore110 */111struct psz_context {112	struct pstore_zone **kpszs;113	struct pstore_zone *ppsz;114	struct pstore_zone *cpsz;115	struct pstore_zone **fpszs;116	unsigned int kmsg_max_cnt;117	unsigned int kmsg_read_cnt;118	unsigned int kmsg_write_cnt;119	unsigned int pmsg_read_cnt;120	unsigned int console_read_cnt;121	unsigned int ftrace_max_cnt;122	unsigned int ftrace_read_cnt;123	/*124	 * These counters should be calculated during recovery.125	 * It records the oops/panic times after crashes rather than boots.126	 */127	unsigned int oops_counter;128	unsigned int panic_counter;129	atomic_t recovered;130	atomic_t on_panic;131 132	/*133	 * pstore_zone_info_lock protects this entire structure during calls134	 * to register_pstore_zone()/unregister_pstore_zone().135	 */136	struct mutex pstore_zone_info_lock;137	struct pstore_zone_info *pstore_zone_info;138	struct pstore_info pstore;139};140static struct psz_context pstore_zone_cxt;141 142static void psz_flush_all_dirty_zones(struct work_struct *);143static DECLARE_DELAYED_WORK(psz_cleaner, psz_flush_all_dirty_zones);144 145/**146 * enum psz_flush_mode - flush mode for psz_zone_write()147 *148 * @FLUSH_NONE: do not flush to storage but update data on memory149 * @FLUSH_PART: just flush part of data including meta data to storage150 * @FLUSH_META: just flush meta data of zone to storage151 * @FLUSH_ALL: flush all of zone152 */153enum psz_flush_mode {154	FLUSH_NONE = 0,155	FLUSH_PART,156	FLUSH_META,157	FLUSH_ALL,158};159 160static inline int buffer_datalen(struct pstore_zone *zone)161{162	return atomic_read(&zone->buffer->datalen);163}164 165static inline int buffer_start(struct pstore_zone *zone)166{167	return atomic_read(&zone->buffer->start);168}169 170static inline bool is_on_panic(void)171{172	return atomic_read(&pstore_zone_cxt.on_panic);173}174 175static ssize_t psz_zone_read_buffer(struct pstore_zone *zone, char *buf,176		size_t len, unsigned long off)177{178	if (!buf || !zone || !zone->buffer)179		return -EINVAL;180	if (off > zone->buffer_size)181		return -EINVAL;182	len = min_t(size_t, len, zone->buffer_size - off);183	memcpy(buf, zone->buffer->data + off, len);184	return len;185}186 187static int psz_zone_read_oldbuf(struct pstore_zone *zone, char *buf,188		size_t len, unsigned long off)189{190	if (!buf || !zone || !zone->oldbuf)191		return -EINVAL;192	if (off > zone->buffer_size)193		return -EINVAL;194	len = min_t(size_t, len, zone->buffer_size - off);195	memcpy(buf, zone->oldbuf->data + off, len);196	return 0;197}198 199static int psz_zone_write(struct pstore_zone *zone,200		enum psz_flush_mode flush_mode, const char *buf,201		size_t len, unsigned long off)202{203	struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info;204	ssize_t wcnt = 0;205	ssize_t (*writeop)(const char *buf, size_t bytes, loff_t pos);206	size_t wlen;207 208	if (off > zone->buffer_size)209		return -EINVAL;210 211	wlen = min_t(size_t, len, zone->buffer_size - off);212	if (buf && wlen) {213		memcpy(zone->buffer->data + off, buf, wlen);214		atomic_set(&zone->buffer->datalen, wlen + off);215	}216 217	/* avoid to damage old records */218	if (!is_on_panic() && !atomic_read(&pstore_zone_cxt.recovered))219		goto dirty;220 221	writeop = is_on_panic() ? info->panic_write : info->write;222	if (!writeop)223		goto dirty;224 225	switch (flush_mode) {226	case FLUSH_NONE:227		if (unlikely(buf && wlen))228			goto dirty;229		return 0;230	case FLUSH_PART:231		wcnt = writeop((const char *)zone->buffer->data + off, wlen,232				zone->off + sizeof(*zone->buffer) + off);233		if (wcnt != wlen)234			goto dirty;235		fallthrough;236	case FLUSH_META:237		wlen = sizeof(struct psz_buffer);238		wcnt = writeop((const char *)zone->buffer, wlen, zone->off);239		if (wcnt != wlen)240			goto dirty;241		break;242	case FLUSH_ALL:243		wlen = zone->buffer_size + sizeof(*zone->buffer);244		wcnt = writeop((const char *)zone->buffer, wlen, zone->off);245		if (wcnt != wlen)246			goto dirty;247		break;248	}249 250	return 0;251dirty:252	/* no need to mark dirty if going to try next zone */253	if (wcnt == -ENOMSG)254		return -ENOMSG;255	atomic_set(&zone->dirty, true);256	/* flush dirty zones nicely */257	if (wcnt == -EBUSY && !is_on_panic())258		schedule_delayed_work(&psz_cleaner, msecs_to_jiffies(500));259	return -EBUSY;260}261 262static int psz_flush_dirty_zone(struct pstore_zone *zone)263{264	int ret;265 266	if (unlikely(!zone))267		return -EINVAL;268 269	if (unlikely(!atomic_read(&pstore_zone_cxt.recovered)))270		return -EBUSY;271 272	if (!atomic_xchg(&zone->dirty, false))273		return 0;274 275	ret = psz_zone_write(zone, FLUSH_ALL, NULL, 0, 0);276	if (ret)277		atomic_set(&zone->dirty, true);278	return ret;279}280 281static int psz_flush_dirty_zones(struct pstore_zone **zones, unsigned int cnt)282{283	int i, ret;284	struct pstore_zone *zone;285 286	if (!zones)287		return -EINVAL;288 289	for (i = 0; i < cnt; i++) {290		zone = zones[i];291		if (!zone)292			return -EINVAL;293		ret = psz_flush_dirty_zone(zone);294		if (ret)295			return ret;296	}297	return 0;298}299 300static int psz_move_zone(struct pstore_zone *old, struct pstore_zone *new)301{302	const char *data = (const char *)old->buffer->data;303	int ret;304 305	ret = psz_zone_write(new, FLUSH_ALL, data, buffer_datalen(old), 0);306	if (ret) {307		atomic_set(&new->buffer->datalen, 0);308		atomic_set(&new->dirty, false);309		return ret;310	}311	atomic_set(&old->buffer->datalen, 0);312	return 0;313}314 315static void psz_flush_all_dirty_zones(struct work_struct *work)316{317	struct psz_context *cxt = &pstore_zone_cxt;318	int ret = 0;319 320	if (cxt->ppsz)321		ret |= psz_flush_dirty_zone(cxt->ppsz);322	if (cxt->cpsz)323		ret |= psz_flush_dirty_zone(cxt->cpsz);324	if (cxt->kpszs)325		ret |= psz_flush_dirty_zones(cxt->kpszs, cxt->kmsg_max_cnt);326	if (cxt->fpszs)327		ret |= psz_flush_dirty_zones(cxt->fpszs, cxt->ftrace_max_cnt);328	if (ret && cxt->pstore_zone_info)329		schedule_delayed_work(&psz_cleaner, msecs_to_jiffies(1000));330}331 332static int psz_kmsg_recover_data(struct psz_context *cxt)333{334	struct pstore_zone_info *info = cxt->pstore_zone_info;335	struct pstore_zone *zone = NULL;336	struct psz_buffer *buf;337	unsigned long i;338	ssize_t rcnt;339 340	if (!info->read)341		return -EINVAL;342 343	for (i = 0; i < cxt->kmsg_max_cnt; i++) {344		zone = cxt->kpszs[i];345		if (unlikely(!zone))346			return -EINVAL;347		if (atomic_read(&zone->dirty)) {348			unsigned int wcnt = cxt->kmsg_write_cnt;349			struct pstore_zone *new = cxt->kpszs[wcnt];350			int ret;351 352			ret = psz_move_zone(zone, new);353			if (ret) {354				pr_err("move zone from %lu to %d failed\n",355						i, wcnt);356				return ret;357			}358			cxt->kmsg_write_cnt = (wcnt + 1) % cxt->kmsg_max_cnt;359		}360		if (!zone->should_recover)361			continue;362		buf = zone->buffer;363		rcnt = info->read((char *)buf, zone->buffer_size + sizeof(*buf),364				zone->off);365		if (rcnt != zone->buffer_size + sizeof(*buf))366			return rcnt < 0 ? rcnt : -EIO;367	}368	return 0;369}370 371static int psz_kmsg_recover_meta(struct psz_context *cxt)372{373	struct pstore_zone_info *info = cxt->pstore_zone_info;374	struct pstore_zone *zone;375	ssize_t rcnt, len;376	struct psz_buffer *buf;377	struct psz_kmsg_header *hdr;378	struct timespec64 time = { };379	unsigned long i;380	/*381	 * Recover may on panic, we can't allocate any memory by kmalloc.382	 * So, we use local array instead.383	 */384	char buffer_header[sizeof(*buf) + sizeof(*hdr)] = {0};385 386	if (!info->read)387		return -EINVAL;388 389	len = sizeof(*buf) + sizeof(*hdr);390	buf = (struct psz_buffer *)buffer_header;391	for (i = 0; i < cxt->kmsg_max_cnt; i++) {392		zone = cxt->kpszs[i];393		if (unlikely(!zone))394			return -EINVAL;395 396		rcnt = info->read((char *)buf, len, zone->off);397		if (rcnt == -ENOMSG) {398			pr_debug("%s with id %lu may be broken, skip\n",399					zone->name, i);400			continue;401		} else if (rcnt != len) {402			pr_err("read %s with id %lu failed\n", zone->name, i);403			return rcnt < 0 ? rcnt : -EIO;404		}405 406		if (buf->sig != zone->buffer->sig) {407			pr_debug("no valid data in kmsg dump zone %lu\n", i);408			continue;409		}410 411		if (zone->buffer_size < atomic_read(&buf->datalen)) {412			pr_info("found overtop zone: %s: id %lu, off %lld, size %zu\n",413					zone->name, i, zone->off,414					zone->buffer_size);415			continue;416		}417 418		hdr = (struct psz_kmsg_header *)buf->data;419		if (hdr->magic != PSTORE_KMSG_HEADER_MAGIC) {420			pr_info("found invalid zone: %s: id %lu, off %lld, size %zu\n",421					zone->name, i, zone->off,422					zone->buffer_size);423			continue;424		}425 426		/*427		 * we get the newest zone, and the next one must be the oldest428		 * or unused zone, because we do write one by one like a circle.429		 */430		if (hdr->time.tv_sec >= time.tv_sec) {431			time.tv_sec = hdr->time.tv_sec;432			cxt->kmsg_write_cnt = (i + 1) % cxt->kmsg_max_cnt;433		}434 435		if (hdr->reason == KMSG_DUMP_OOPS)436			cxt->oops_counter =437				max(cxt->oops_counter, hdr->counter);438		else if (hdr->reason == KMSG_DUMP_PANIC)439			cxt->panic_counter =440				max(cxt->panic_counter, hdr->counter);441 442		if (!atomic_read(&buf->datalen)) {443			pr_debug("found erased zone: %s: id %lu, off %lld, size %zu, datalen %d\n",444					zone->name, i, zone->off,445					zone->buffer_size,446					atomic_read(&buf->datalen));447			continue;448		}449 450		if (!is_on_panic())451			zone->should_recover = true;452		pr_debug("found nice zone: %s: id %lu, off %lld, size %zu, datalen %d\n",453				zone->name, i, zone->off,454				zone->buffer_size, atomic_read(&buf->datalen));455	}456 457	return 0;458}459 460static int psz_kmsg_recover(struct psz_context *cxt)461{462	int ret;463 464	if (!cxt->kpszs)465		return 0;466 467	ret = psz_kmsg_recover_meta(cxt);468	if (ret)469		goto recover_fail;470 471	ret = psz_kmsg_recover_data(cxt);472	if (ret)473		goto recover_fail;474 475	return 0;476recover_fail:477	pr_debug("psz_recover_kmsg failed\n");478	return ret;479}480 481static int psz_recover_zone(struct psz_context *cxt, struct pstore_zone *zone)482{483	struct pstore_zone_info *info = cxt->pstore_zone_info;484	struct psz_buffer *oldbuf, tmpbuf;485	int ret = 0;486	char *buf;487	ssize_t rcnt, len, start, off;488 489	if (!zone || zone->oldbuf)490		return 0;491 492	if (is_on_panic()) {493		/* save data as much as possible */494		psz_flush_dirty_zone(zone);495		return 0;496	}497 498	if (unlikely(!info->read))499		return -EINVAL;500 501	len = sizeof(struct psz_buffer);502	rcnt = info->read((char *)&tmpbuf, len, zone->off);503	if (rcnt != len) {504		pr_debug("read zone %s failed\n", zone->name);505		return rcnt < 0 ? rcnt : -EIO;506	}507 508	if (tmpbuf.sig != zone->buffer->sig) {509		pr_debug("no valid data in zone %s\n", zone->name);510		return 0;511	}512 513	if (zone->buffer_size < atomic_read(&tmpbuf.datalen) ||514		zone->buffer_size < atomic_read(&tmpbuf.start)) {515		pr_info("found overtop zone: %s: off %lld, size %zu\n",516				zone->name, zone->off, zone->buffer_size);517		/* just keep going */518		return 0;519	}520 521	if (!atomic_read(&tmpbuf.datalen)) {522		pr_debug("found erased zone: %s: off %lld, size %zu, datalen %d\n",523				zone->name, zone->off, zone->buffer_size,524				atomic_read(&tmpbuf.datalen));525		return 0;526	}527 528	pr_debug("found nice zone: %s: off %lld, size %zu, datalen %d\n",529			zone->name, zone->off, zone->buffer_size,530			atomic_read(&tmpbuf.datalen));531 532	len = atomic_read(&tmpbuf.datalen) + sizeof(*oldbuf);533	oldbuf = kzalloc(len, GFP_KERNEL);534	if (!oldbuf)535		return -ENOMEM;536 537	memcpy(oldbuf, &tmpbuf, sizeof(*oldbuf));538	buf = (char *)oldbuf + sizeof(*oldbuf);539	len = atomic_read(&oldbuf->datalen);540	start = atomic_read(&oldbuf->start);541	off = zone->off + sizeof(*oldbuf);542 543	/* get part of data */544	rcnt = info->read(buf, len - start, off + start);545	if (rcnt != len - start) {546		pr_err("read zone %s failed\n", zone->name);547		ret = rcnt < 0 ? rcnt : -EIO;548		goto free_oldbuf;549	}550 551	/* get the rest of data */552	rcnt = info->read(buf + len - start, start, off);553	if (rcnt != start) {554		pr_err("read zone %s failed\n", zone->name);555		ret = rcnt < 0 ? rcnt : -EIO;556		goto free_oldbuf;557	}558 559	zone->oldbuf = oldbuf;560	psz_flush_dirty_zone(zone);561	return 0;562 563free_oldbuf:564	kfree(oldbuf);565	return ret;566}567 568static int psz_recover_zones(struct psz_context *cxt,569		struct pstore_zone **zones, unsigned int cnt)570{571	int ret;572	unsigned int i;573	struct pstore_zone *zone;574 575	if (!zones)576		return 0;577 578	for (i = 0; i < cnt; i++) {579		zone = zones[i];580		if (unlikely(!zone))581			continue;582		ret = psz_recover_zone(cxt, zone);583		if (ret)584			goto recover_fail;585	}586 587	return 0;588recover_fail:589	pr_debug("recover %s[%u] failed\n", zone->name, i);590	return ret;591}592 593/**594 * psz_recovery() - recover data from storage595 * @cxt: the context of pstore/zone596 *597 * recovery means reading data back from storage after rebooting598 *599 * Return: 0 on success, others on failure.600 */601static inline int psz_recovery(struct psz_context *cxt)602{603	int ret;604 605	if (atomic_read(&cxt->recovered))606		return 0;607 608	ret = psz_kmsg_recover(cxt);609	if (ret)610		goto out;611 612	ret = psz_recover_zone(cxt, cxt->ppsz);613	if (ret)614		goto out;615 616	ret = psz_recover_zone(cxt, cxt->cpsz);617	if (ret)618		goto out;619 620	ret = psz_recover_zones(cxt, cxt->fpszs, cxt->ftrace_max_cnt);621 622out:623	if (unlikely(ret))624		pr_err("recover failed\n");625	else {626		pr_debug("recover end!\n");627		atomic_set(&cxt->recovered, 1);628	}629	return ret;630}631 632static int psz_pstore_open(struct pstore_info *psi)633{634	struct psz_context *cxt = psi->data;635 636	cxt->kmsg_read_cnt = 0;637	cxt->pmsg_read_cnt = 0;638	cxt->console_read_cnt = 0;639	cxt->ftrace_read_cnt = 0;640	return 0;641}642 643static inline bool psz_old_ok(struct pstore_zone *zone)644{645	if (zone && zone->oldbuf && atomic_read(&zone->oldbuf->datalen))646		return true;647	return false;648}649 650static inline bool psz_ok(struct pstore_zone *zone)651{652	if (zone && zone->buffer && buffer_datalen(zone))653		return true;654	return false;655}656 657static inline int psz_kmsg_erase(struct psz_context *cxt,658		struct pstore_zone *zone, struct pstore_record *record)659{660	struct psz_buffer *buffer = zone->buffer;661	struct psz_kmsg_header *hdr =662		(struct psz_kmsg_header *)buffer->data;663	size_t size;664 665	if (unlikely(!psz_ok(zone)))666		return 0;667 668	/* this zone is already updated, no need to erase */669	if (record->count != hdr->counter)670		return 0;671 672	size = buffer_datalen(zone) + sizeof(*zone->buffer);673	atomic_set(&zone->buffer->datalen, 0);674	if (cxt->pstore_zone_info->erase)675		return cxt->pstore_zone_info->erase(size, zone->off);676	else677		return psz_zone_write(zone, FLUSH_META, NULL, 0, 0);678}679 680static inline int psz_record_erase(struct psz_context *cxt,681		struct pstore_zone *zone)682{683	if (unlikely(!psz_old_ok(zone)))684		return 0;685 686	kfree(zone->oldbuf);687	zone->oldbuf = NULL;688	/*689	 * if there are new data in zone buffer, that means the old data690	 * are already invalid. It is no need to flush 0 (erase) to691	 * block device.692	 */693	if (!buffer_datalen(zone))694		return psz_zone_write(zone, FLUSH_META, NULL, 0, 0);695	psz_flush_dirty_zone(zone);696	return 0;697}698 699static int psz_pstore_erase(struct pstore_record *record)700{701	struct psz_context *cxt = record->psi->data;702 703	switch (record->type) {704	case PSTORE_TYPE_DMESG:705		if (record->id >= cxt->kmsg_max_cnt)706			return -EINVAL;707		return psz_kmsg_erase(cxt, cxt->kpszs[record->id], record);708	case PSTORE_TYPE_PMSG:709		return psz_record_erase(cxt, cxt->ppsz);710	case PSTORE_TYPE_CONSOLE:711		return psz_record_erase(cxt, cxt->cpsz);712	case PSTORE_TYPE_FTRACE:713		if (record->id >= cxt->ftrace_max_cnt)714			return -EINVAL;715		return psz_record_erase(cxt, cxt->fpszs[record->id]);716	default: return -EINVAL;717	}718}719 720static void psz_write_kmsg_hdr(struct pstore_zone *zone,721		struct pstore_record *record)722{723	struct psz_context *cxt = record->psi->data;724	struct psz_buffer *buffer = zone->buffer;725	struct psz_kmsg_header *hdr =726		(struct psz_kmsg_header *)buffer->data;727 728	hdr->magic = PSTORE_KMSG_HEADER_MAGIC;729	hdr->compressed = record->compressed;730	hdr->time.tv_sec = record->time.tv_sec;731	hdr->time.tv_nsec = record->time.tv_nsec;732	hdr->reason = record->reason;733	if (hdr->reason == KMSG_DUMP_OOPS)734		hdr->counter = ++cxt->oops_counter;735	else if (hdr->reason == KMSG_DUMP_PANIC)736		hdr->counter = ++cxt->panic_counter;737	else738		hdr->counter = 0;739}740 741/*742 * In case zone is broken, which may occur to MTD device, we try each zones,743 * start at cxt->kmsg_write_cnt.744 */745static inline int notrace psz_kmsg_write_record(struct psz_context *cxt,746		struct pstore_record *record)747{748	size_t size, hlen;749	struct pstore_zone *zone;750	unsigned int i;751 752	for (i = 0; i < cxt->kmsg_max_cnt; i++) {753		unsigned int zonenum, len;754		int ret;755 756		zonenum = (cxt->kmsg_write_cnt + i) % cxt->kmsg_max_cnt;757		zone = cxt->kpszs[zonenum];758		if (unlikely(!zone))759			return -ENOSPC;760 761		/* avoid destroying old data, allocate a new one */762		len = zone->buffer_size + sizeof(*zone->buffer);763		zone->oldbuf = zone->buffer;764		zone->buffer = kzalloc(len, GFP_ATOMIC);765		if (!zone->buffer) {766			zone->buffer = zone->oldbuf;767			return -ENOMEM;768		}769		zone->buffer->sig = zone->oldbuf->sig;770 771		pr_debug("write %s to zone id %d\n", zone->name, zonenum);772		psz_write_kmsg_hdr(zone, record);773		hlen = sizeof(struct psz_kmsg_header);774		size = min_t(size_t, record->size, zone->buffer_size - hlen);775		ret = psz_zone_write(zone, FLUSH_ALL, record->buf, size, hlen);776		if (likely(!ret || ret != -ENOMSG)) {777			cxt->kmsg_write_cnt = zonenum + 1;778			cxt->kmsg_write_cnt %= cxt->kmsg_max_cnt;779			/* no need to try next zone, free last zone buffer */780			kfree(zone->oldbuf);781			zone->oldbuf = NULL;782			return ret;783		}784 785		pr_debug("zone %u may be broken, try next dmesg zone\n",786				zonenum);787		kfree(zone->buffer);788		zone->buffer = zone->oldbuf;789		zone->oldbuf = NULL;790	}791 792	return -EBUSY;793}794 795static int notrace psz_kmsg_write(struct psz_context *cxt,796		struct pstore_record *record)797{798	int ret;799 800	/*801	 * Explicitly only take the first part of any new crash.802	 * If our buffer is larger than kmsg_bytes, this can never happen,803	 * and if our buffer is smaller than kmsg_bytes, we don't want the804	 * report split across multiple records.805	 */806	if (record->part != 1)807		return -ENOSPC;808 809	if (!cxt->kpszs)810		return -ENOSPC;811 812	ret = psz_kmsg_write_record(cxt, record);813	if (!ret && is_on_panic()) {814		/* ensure all data are flushed to storage when panic */815		pr_debug("try to flush other dirty zones\n");816		psz_flush_all_dirty_zones(NULL);817	}818 819	/* always return 0 as we had handled it on buffer */820	return 0;821}822 823static int notrace psz_record_write(struct pstore_zone *zone,824		struct pstore_record *record)825{826	size_t start, rem;827	bool is_full_data = false;828	char *buf;829	int cnt;830 831	if (!zone || !record)832		return -ENOSPC;833 834	if (atomic_read(&zone->buffer->datalen) >= zone->buffer_size)835		is_full_data = true;836 837	cnt = record->size;838	buf = record->buf;839	if (unlikely(cnt > zone->buffer_size)) {840		buf += cnt - zone->buffer_size;841		cnt = zone->buffer_size;842	}843 844	start = buffer_start(zone);845	rem = zone->buffer_size - start;846	if (unlikely(rem < cnt)) {847		psz_zone_write(zone, FLUSH_PART, buf, rem, start);848		buf += rem;849		cnt -= rem;850		start = 0;851		is_full_data = true;852	}853 854	atomic_set(&zone->buffer->start, cnt + start);855	psz_zone_write(zone, FLUSH_PART, buf, cnt, start);856 857	/**858	 * psz_zone_write will set datalen as start + cnt.859	 * It work if actual data length lesser than buffer size.860	 * If data length greater than buffer size, pmsg will rewrite to861	 * beginning of zone, which make buffer->datalen wrongly.862	 * So we should reset datalen as buffer size once actual data length863	 * greater than buffer size.864	 */865	if (is_full_data) {866		atomic_set(&zone->buffer->datalen, zone->buffer_size);867		psz_zone_write(zone, FLUSH_META, NULL, 0, 0);868	}869	return 0;870}871 872static int notrace psz_pstore_write(struct pstore_record *record)873{874	struct psz_context *cxt = record->psi->data;875 876	if (record->type == PSTORE_TYPE_DMESG &&877			record->reason == KMSG_DUMP_PANIC)878		atomic_set(&cxt->on_panic, 1);879 880	/*881	 * if on panic, do not write except panic records882	 * Fix case that panic_write prints log which wakes up console backend.883	 */884	if (is_on_panic() && record->type != PSTORE_TYPE_DMESG)885		return -EBUSY;886 887	switch (record->type) {888	case PSTORE_TYPE_DMESG:889		return psz_kmsg_write(cxt, record);890	case PSTORE_TYPE_CONSOLE:891		return psz_record_write(cxt->cpsz, record);892	case PSTORE_TYPE_PMSG:893		return psz_record_write(cxt->ppsz, record);894	case PSTORE_TYPE_FTRACE: {895		int zonenum = smp_processor_id();896 897		if (!cxt->fpszs)898			return -ENOSPC;899		return psz_record_write(cxt->fpszs[zonenum], record);900	}901	default:902		return -EINVAL;903	}904}905 906static struct pstore_zone *psz_read_next_zone(struct psz_context *cxt)907{908	struct pstore_zone *zone = NULL;909 910	while (cxt->kmsg_read_cnt < cxt->kmsg_max_cnt) {911		zone = cxt->kpszs[cxt->kmsg_read_cnt++];912		if (psz_ok(zone))913			return zone;914	}915 916	if (cxt->ftrace_read_cnt < cxt->ftrace_max_cnt)917		/*918		 * No need psz_old_ok(). Let psz_ftrace_read() do so for919		 * combination. psz_ftrace_read() should traverse over920		 * all zones in case of some zone without data.921		 */922		return cxt->fpszs[cxt->ftrace_read_cnt++];923 924	if (cxt->pmsg_read_cnt == 0) {925		cxt->pmsg_read_cnt++;926		zone = cxt->ppsz;927		if (psz_old_ok(zone))928			return zone;929	}930 931	if (cxt->console_read_cnt == 0) {932		cxt->console_read_cnt++;933		zone = cxt->cpsz;934		if (psz_old_ok(zone))935			return zone;936	}937 938	return NULL;939}940 941static int psz_kmsg_read_hdr(struct pstore_zone *zone,942		struct pstore_record *record)943{944	struct psz_buffer *buffer = zone->buffer;945	struct psz_kmsg_header *hdr =946		(struct psz_kmsg_header *)buffer->data;947 948	if (hdr->magic != PSTORE_KMSG_HEADER_MAGIC)949		return -EINVAL;950	record->compressed = hdr->compressed;951	record->time.tv_sec = hdr->time.tv_sec;952	record->time.tv_nsec = hdr->time.tv_nsec;953	record->reason = hdr->reason;954	record->count = hdr->counter;955	return 0;956}957 958static ssize_t psz_kmsg_read(struct pstore_zone *zone,959		struct pstore_record *record)960{961	ssize_t size, hlen = 0;962 963	size = buffer_datalen(zone);964	/* Clear and skip this kmsg dump record if it has no valid header */965	if (psz_kmsg_read_hdr(zone, record)) {966		atomic_set(&zone->buffer->datalen, 0);967		atomic_set(&zone->dirty, 0);968		return -ENOMSG;969	}970	size -= sizeof(struct psz_kmsg_header);971 972	if (!record->compressed) {973		char *buf = kasprintf(GFP_KERNEL, "%s: Total %d times\n",974				      kmsg_dump_reason_str(record->reason),975				      record->count);976		if (!buf)977			return -ENOMEM;978		hlen = strlen(buf);979		record->buf = krealloc(buf, hlen + size, GFP_KERNEL);980		if (!record->buf) {981			kfree(buf);982			return -ENOMEM;983		}984	} else {985		record->buf = kmalloc(size, GFP_KERNEL);986		if (!record->buf)987			return -ENOMEM;988	}989 990	size = psz_zone_read_buffer(zone, record->buf + hlen, size,991			sizeof(struct psz_kmsg_header));992	if (unlikely(size < 0)) {993		kfree(record->buf);994		return -ENOMSG;995	}996 997	return size + hlen;998}999 1000/* try to combine all ftrace zones */1001static ssize_t psz_ftrace_read(struct pstore_zone *zone,1002		struct pstore_record *record)1003{1004	struct psz_context *cxt;1005	struct psz_buffer *buf;1006	int ret;1007 1008	if (!zone || !record)1009		return -ENOSPC;1010 1011	if (!psz_old_ok(zone))1012		goto out;1013 1014	buf = (struct psz_buffer *)zone->oldbuf;1015	if (!buf)1016		return -ENOMSG;1017 1018	ret = pstore_ftrace_combine_log(&record->buf, &record->size,1019			(char *)buf->data, atomic_read(&buf->datalen));1020	if (unlikely(ret))1021		return ret;1022 1023out:1024	cxt = record->psi->data;1025	if (cxt->ftrace_read_cnt < cxt->ftrace_max_cnt)1026		/* then, read next ftrace zone */1027		return -ENOMSG;1028	record->id = 0;1029	return record->size ? record->size : -ENOMSG;1030}1031 1032static ssize_t psz_record_read(struct pstore_zone *zone,1033		struct pstore_record *record)1034{1035	size_t len;1036	struct psz_buffer *buf;1037 1038	if (!zone || !record)1039		return -ENOSPC;1040 1041	buf = (struct psz_buffer *)zone->oldbuf;1042	if (!buf)1043		return -ENOMSG;1044 1045	len = atomic_read(&buf->datalen);1046	record->buf = kmalloc(len, GFP_KERNEL);1047	if (!record->buf)1048		return -ENOMEM;1049 1050	if (unlikely(psz_zone_read_oldbuf(zone, record->buf, len, 0))) {1051		kfree(record->buf);1052		return -ENOMSG;1053	}1054 1055	return len;1056}1057 1058static ssize_t psz_pstore_read(struct pstore_record *record)1059{1060	struct psz_context *cxt = record->psi->data;1061	ssize_t (*readop)(struct pstore_zone *zone,1062			struct pstore_record *record);1063	struct pstore_zone *zone;1064	ssize_t ret;1065 1066	/* before read, we must recover from storage */1067	ret = psz_recovery(cxt);1068	if (ret)1069		return ret;1070 1071next_zone:1072	zone = psz_read_next_zone(cxt);1073	if (!zone)1074		return 0;1075 1076	record->type = zone->type;1077	switch (record->type) {1078	case PSTORE_TYPE_DMESG:1079		readop = psz_kmsg_read;1080		record->id = cxt->kmsg_read_cnt - 1;1081		break;1082	case PSTORE_TYPE_FTRACE:1083		readop = psz_ftrace_read;1084		break;1085	case PSTORE_TYPE_CONSOLE:1086	case PSTORE_TYPE_PMSG:1087		readop = psz_record_read;1088		break;1089	default:1090		goto next_zone;1091	}1092 1093	ret = readop(zone, record);1094	if (ret == -ENOMSG)1095		goto next_zone;1096	return ret;1097}1098 1099static struct psz_context pstore_zone_cxt = {1100	.pstore_zone_info_lock =1101		__MUTEX_INITIALIZER(pstore_zone_cxt.pstore_zone_info_lock),1102	.recovered = ATOMIC_INIT(0),1103	.on_panic = ATOMIC_INIT(0),1104	.pstore = {1105		.owner = THIS_MODULE,1106		.open = psz_pstore_open,1107		.read = psz_pstore_read,1108		.write = psz_pstore_write,1109		.erase = psz_pstore_erase,1110	},1111};1112 1113static void psz_free_zone(struct pstore_zone **pszone)1114{1115	struct pstore_zone *zone = *pszone;1116 1117	if (!zone)1118		return;1119 1120	kfree(zone->buffer);1121	kfree(zone);1122	*pszone = NULL;1123}1124 1125static void psz_free_zones(struct pstore_zone ***pszones, unsigned int *cnt)1126{1127	struct pstore_zone **zones = *pszones;1128 1129	if (!zones)1130		return;1131 1132	while (*cnt > 0) {1133		(*cnt)--;1134		psz_free_zone(&(zones[*cnt]));1135	}1136	kfree(zones);1137	*pszones = NULL;1138}1139 1140static void psz_free_all_zones(struct psz_context *cxt)1141{1142	if (cxt->kpszs)1143		psz_free_zones(&cxt->kpszs, &cxt->kmsg_max_cnt);1144	if (cxt->ppsz)1145		psz_free_zone(&cxt->ppsz);1146	if (cxt->cpsz)1147		psz_free_zone(&cxt->cpsz);1148	if (cxt->fpszs)1149		psz_free_zones(&cxt->fpszs, &cxt->ftrace_max_cnt);1150}1151 1152static struct pstore_zone *psz_init_zone(enum pstore_type_id type,1153		loff_t *off, size_t size)1154{1155	struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info;1156	struct pstore_zone *zone;1157	const char *name = pstore_type_to_name(type);1158 1159	if (!size)1160		return NULL;1161 1162	if (*off + size > info->total_size) {1163		pr_err("no room for %s (0x%zx@0x%llx over 0x%lx)\n",1164			name, size, *off, info->total_size);1165		return ERR_PTR(-ENOMEM);1166	}1167 1168	zone = kzalloc(sizeof(struct pstore_zone), GFP_KERNEL);1169	if (!zone)1170		return ERR_PTR(-ENOMEM);1171 1172	zone->buffer = kmalloc(size, GFP_KERNEL);1173	if (!zone->buffer) {1174		kfree(zone);1175		return ERR_PTR(-ENOMEM);1176	}1177	memset(zone->buffer, 0xFF, size);1178	zone->off = *off;1179	zone->name = name;1180	zone->type = type;1181	zone->buffer_size = size - sizeof(struct psz_buffer);1182	zone->buffer->sig = type ^ PSZ_SIG;1183	zone->oldbuf = NULL;1184	atomic_set(&zone->dirty, 0);1185	atomic_set(&zone->buffer->datalen, 0);1186	atomic_set(&zone->buffer->start, 0);1187 1188	*off += size;1189 1190	pr_debug("pszone %s: off 0x%llx, %zu header, %zu data\n", zone->name,1191			zone->off, sizeof(*zone->buffer), zone->buffer_size);1192	return zone;1193}1194 1195static struct pstore_zone **psz_init_zones(enum pstore_type_id type,1196	loff_t *off, size_t total_size, ssize_t record_size,1197	unsigned int *cnt)1198{1199	struct pstore_zone_info *info = pstore_zone_cxt.pstore_zone_info;1200	struct pstore_zone **zones, *zone;1201	const char *name = pstore_type_to_name(type);1202	int c, i;1203 1204	*cnt = 0;1205	if (!total_size || !record_size)1206		return NULL;1207 1208	if (*off + total_size > info->total_size) {1209		pr_err("no room for zones %s (0x%zx@0x%llx over 0x%lx)\n",1210			name, total_size, *off, info->total_size);1211		return ERR_PTR(-ENOMEM);1212	}1213 1214	c = total_size / record_size;1215	zones = kcalloc(c, sizeof(*zones), GFP_KERNEL);1216	if (!zones) {1217		pr_err("allocate for zones %s failed\n", name);1218		return ERR_PTR(-ENOMEM);1219	}1220 1221	for (i = 0; i < c; i++) {1222		zone = psz_init_zone(type, off, record_size);1223		if (!zone || IS_ERR(zone)) {1224			pr_err("initialize zones %s failed\n", name);1225			psz_free_zones(&zones, &i);1226			return (void *)zone;1227		}1228		zones[i] = zone;1229	}1230 1231	*cnt = c;1232	return zones;1233}1234 1235static int psz_alloc_zones(struct psz_context *cxt)1236{1237	struct pstore_zone_info *info = cxt->pstore_zone_info;1238	loff_t off = 0;1239	int err;1240	size_t off_size = 0;1241 1242	off_size += info->pmsg_size;1243	cxt->ppsz = psz_init_zone(PSTORE_TYPE_PMSG, &off, info->pmsg_size);1244	if (IS_ERR(cxt->ppsz)) {1245		err = PTR_ERR(cxt->ppsz);1246		cxt->ppsz = NULL;1247		goto free_out;1248	}1249 1250	off_size += info->console_size;1251	cxt->cpsz = psz_init_zone(PSTORE_TYPE_CONSOLE, &off,1252			info->console_size);1253	if (IS_ERR(cxt->cpsz)) {1254		err = PTR_ERR(cxt->cpsz);1255		cxt->cpsz = NULL;1256		goto free_out;1257	}1258 1259	off_size += info->ftrace_size;1260	cxt->fpszs = psz_init_zones(PSTORE_TYPE_FTRACE, &off,1261			info->ftrace_size,1262			info->ftrace_size / nr_cpu_ids,1263			&cxt->ftrace_max_cnt);1264	if (IS_ERR(cxt->fpszs)) {1265		err = PTR_ERR(cxt->fpszs);1266		cxt->fpszs = NULL;1267		goto free_out;1268	}1269 1270	cxt->kpszs = psz_init_zones(PSTORE_TYPE_DMESG, &off,1271			info->total_size - off_size,1272			info->kmsg_size, &cxt->kmsg_max_cnt);1273	if (IS_ERR(cxt->kpszs)) {1274		err = PTR_ERR(cxt->kpszs);1275		cxt->kpszs = NULL;1276		goto free_out;1277	}1278 1279	return 0;1280free_out:1281	psz_free_all_zones(cxt);1282	return err;1283}1284 1285/**1286 * register_pstore_zone() - register to pstore/zone1287 *1288 * @info: back-end driver information. See &struct pstore_zone_info.1289 *1290 * Only one back-end at one time.1291 *1292 * Return: 0 on success, others on failure.1293 */1294int register_pstore_zone(struct pstore_zone_info *info)1295{1296	int err = -EINVAL;1297	struct psz_context *cxt = &pstore_zone_cxt;1298 1299	if (info->total_size < 4096) {1300		pr_warn("total_size must be >= 4096\n");1301		return -EINVAL;1302	}1303	if (info->total_size > SZ_128M) {1304		pr_warn("capping size to 128MiB\n");1305		info->total_size = SZ_128M;1306	}1307 1308	if (!info->kmsg_size && !info->pmsg_size && !info->console_size &&1309	    !info->ftrace_size) {1310		pr_warn("at least one record size must be non-zero\n");1311		return -EINVAL;1312	}1313 1314	if (!info->name || !info->name[0])1315		return -EINVAL;1316 1317#define check_size(name, size) {					\1318		if (info->name > 0 && info->name < (size)) {		\1319			pr_err(#name " must be over %d\n", (size));	\1320			return -EINVAL;					\1321		}							\1322		if (info->name & (size - 1)) {				\1323			pr_err(#name " must be a multiple of %d\n",	\1324					(size));			\1325			return -EINVAL;					\1326		}							\1327	}1328 1329	check_size(total_size, 4096);1330	check_size(kmsg_size, SECTOR_SIZE);1331	check_size(pmsg_size, SECTOR_SIZE);1332	check_size(console_size, SECTOR_SIZE);1333	check_size(ftrace_size, SECTOR_SIZE);1334 1335#undef check_size1336 1337	/*1338	 * the @read and @write must be applied.1339	 * if no @read, pstore may mount failed.1340	 * if no @write, pstore do not support to remove record file.1341	 */1342	if (!info->read || !info->write) {1343		pr_err("no valid general read/write interface\n");1344		return -EINVAL;1345	}1346 1347	mutex_lock(&cxt->pstore_zone_info_lock);1348	if (cxt->pstore_zone_info) {1349		pr_warn("'%s' already loaded: ignoring '%s'\n",1350				cxt->pstore_zone_info->name, info->name);1351		mutex_unlock(&cxt->pstore_zone_info_lock);1352		return -EBUSY;1353	}1354	cxt->pstore_zone_info = info;1355 1356	pr_debug("register %s with properties:\n", info->name);1357	pr_debug("\ttotal size : %ld Bytes\n", info->total_size);1358	pr_debug("\tkmsg size : %ld Bytes\n", info->kmsg_size);1359	pr_debug("\tpmsg size : %ld Bytes\n", info->pmsg_size);1360	pr_debug("\tconsole size : %ld Bytes\n", info->console_size);1361	pr_debug("\tftrace size : %ld Bytes\n", info->ftrace_size);1362 1363	err = psz_alloc_zones(cxt);1364	if (err) {1365		pr_err("alloc zones failed\n");1366		goto fail_out;1367	}1368 1369	if (info->kmsg_size) {1370		cxt->pstore.bufsize = cxt->kpszs[0]->buffer_size -1371			sizeof(struct psz_kmsg_header);1372		cxt->pstore.buf = kzalloc(cxt->pstore.bufsize, GFP_KERNEL);1373		if (!cxt->pstore.buf) {1374			err = -ENOMEM;1375			goto fail_free;1376		}1377	}1378	cxt->pstore.data = cxt;1379 1380	pr_info("registered %s as backend for", info->name);1381	cxt->pstore.max_reason = info->max_reason;1382	cxt->pstore.name = info->name;1383	if (info->kmsg_size) {1384		cxt->pstore.flags |= PSTORE_FLAGS_DMESG;1385		pr_cont(" kmsg(%s",1386			kmsg_dump_reason_str(cxt->pstore.max_reason));1387		if (cxt->pstore_zone_info->panic_write)1388			pr_cont(",panic_write");1389		pr_cont(")");1390	}1391	if (info->pmsg_size) {1392		cxt->pstore.flags |= PSTORE_FLAGS_PMSG;1393		pr_cont(" pmsg");1394	}1395	if (info->console_size) {1396		cxt->pstore.flags |= PSTORE_FLAGS_CONSOLE;1397		pr_cont(" console");1398	}1399	if (info->ftrace_size) {1400		cxt->pstore.flags |= PSTORE_FLAGS_FTRACE;1401		pr_cont(" ftrace");1402	}1403	pr_cont("\n");1404 1405	err = pstore_register(&cxt->pstore);1406	if (err) {1407		pr_err("registering with pstore failed\n");1408		goto fail_free;1409	}1410	mutex_unlock(&pstore_zone_cxt.pstore_zone_info_lock);1411 1412	return 0;1413 1414fail_free:1415	kfree(cxt->pstore.buf);1416	cxt->pstore.buf = NULL;1417	cxt->pstore.bufsize = 0;1418	psz_free_all_zones(cxt);1419fail_out:1420	pstore_zone_cxt.pstore_zone_info = NULL;1421	mutex_unlock(&pstore_zone_cxt.pstore_zone_info_lock);1422	return err;1423}1424EXPORT_SYMBOL_GPL(register_pstore_zone);1425 1426/**1427 * unregister_pstore_zone() - unregister to pstore/zone1428 *1429 * @info: back-end driver information. See struct pstore_zone_info.1430 */1431void unregister_pstore_zone(struct pstore_zone_info *info)1432{1433	struct psz_context *cxt = &pstore_zone_cxt;1434 1435	mutex_lock(&cxt->pstore_zone_info_lock);1436	if (!cxt->pstore_zone_info) {1437		mutex_unlock(&cxt->pstore_zone_info_lock);1438		return;1439	}1440 1441	/* Stop incoming writes from pstore. */1442	pstore_unregister(&cxt->pstore);1443 1444	/* Flush any pending writes. */1445	psz_flush_all_dirty_zones(NULL);1446	flush_delayed_work(&psz_cleaner);1447 1448	/* Clean up allocations. */1449	kfree(cxt->pstore.buf);1450	cxt->pstore.buf = NULL;1451	cxt->pstore.bufsize = 0;1452	cxt->pstore_zone_info = NULL;1453 1454	psz_free_all_zones(cxt);1455 1456	/* Clear counters and zone state. */1457	cxt->oops_counter = 0;1458	cxt->panic_counter = 0;1459	atomic_set(&cxt->recovered, 0);1460	atomic_set(&cxt->on_panic, 0);1461 1462	mutex_unlock(&cxt->pstore_zone_info_lock);1463}1464EXPORT_SYMBOL_GPL(unregister_pstore_zone);1465 1466MODULE_LICENSE("GPL");1467MODULE_AUTHOR("WeiXiong Liao <liaoweixiong@allwinnertech.com>");1468MODULE_AUTHOR("Kees Cook <keescook@chromium.org>");1469MODULE_DESCRIPTION("Storage Manager for pstore/blk");1470